{
 "license": "each record keeps its source licence (source_licence); see /data/LICENSES.md",
 "attribution": "Aider polyglot leaderboard data (Apache-2.0); Arena leaderboard dataset on Hugging Face (CC BY 4.0); Berkeley Function Calling Leaderboard, Gorilla project (Apache-2.0); Epoch AI, ‘Capabilities & benchmarking’ (CC BY 4.0); LiteLLM, model_prices_and_context_window.json (MIT); Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.",
 "field_notes": {
  "source_url": "where the source file is downloaded (for Epoch AI the archive's download URL, whose bytes change on every download)",
  "source_snapshot": "the exact file the value was read from — an immutable dated file, a GitHub commit, or the Epoch AI archive with '#<member file>@sha256:<archive SHA-256 prefix>' — whose SHA-256 is source_sha256",
  "source_sha256": "SHA-256 of the file named by source_snapshot",
  "source_value_text": "the value exactly as the source writes it",
  "retrieved_at": "when this value was first retrieved from the source (it is kept while the source keeps showing the same value)",
  "eval_date": "the evaluation date the source gives for the result"
 },
 "scores": [
  {
   "model": "gpt-6-astra",
   "benchmark": "chess_puzzles",
   "score": 0.74,
   "source_value_text": "0.74",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "furniture_assembly",
   "score": 0.8,
   "source_value_text": "0.8",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "mystery_game_puzzles",
   "score": 0.84,
   "source_value_text": "0.84",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "frontiermath_t4",
   "score": 0.976,
   "source_value_text": "0.976",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "frontiermath_t13",
   "score": 0.9368421052631579,
   "source_value_text": "0.9368421052631579",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "gpqa_diamond",
   "score": 0.9577020202020202,
   "source_value_text": "0.9577020202020202",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9472503333333333,
   "source_value_text": "0.9472503333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "simpleqa_verified",
   "score": 0.756,
   "source_value_text": "0.756",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gpt-6-astra",
   "benchmark": "or_tau2_airline",
   "score": 0.6766665000000001,
   "source_value_text": "0.6766665000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "chess_puzzles",
   "score": 0.68,
   "source_value_text": "0.68",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "furniture_assembly",
   "score": 0.8,
   "source_value_text": "0.8",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "mystery_game_puzzles",
   "score": 0.8,
   "source_value_text": "0.8",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "frontiermath_t4",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "frontiermath_t13",
   "score": 0.9368421052631579,
   "source_value_text": "0.9368421052631579",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "gpqa_diamond",
   "score": 0.9539141414141414,
   "source_value_text": "0.9539141414141414",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "or_gpqa_diamond",
   "score": 0.944444,
   "source_value_text": "0.944444",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "simpleqa_verified",
   "score": 0.739,
   "source_value_text": "0.739",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "gpt-6-1-sol",
   "benchmark": "or_tau2_airline",
   "score": 0.6822223333333333,
   "source_value_text": "0.6822223333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "chess_puzzles",
   "score": 0.67,
   "source_value_text": "0.67",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "furniture_assembly",
   "score": 0.8333333333333334,
   "source_value_text": "0.8333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.71,
   "source_value_text": "0.71",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "frontiermath_t4",
   "score": 0.95,
   "source_value_text": "0.95",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "frontiermath_t13",
   "score": 0.9122807017543859,
   "source_value_text": "0.9122807017543859",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "gpqa_diamond",
   "score": 0.9059343434343434,
   "source_value_text": "0.9059343434343434",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9020199999999999,
   "source_value_text": "0.9020199999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "simpleqa_verified",
   "score": 0.722,
   "source_value_text": "0.722",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "claude-opus-5-5",
   "benchmark": "or_tau2_airline",
   "score": 0.756,
   "source_value_text": "0.756",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "chess_puzzles",
   "score": 0.59,
   "source_value_text": "0.59",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "furniture_assembly",
   "score": 0.75,
   "source_value_text": "0.75",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.65,
   "source_value_text": "0.65",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "frontiermath_t4",
   "score": 0.8048780487804879,
   "source_value_text": "0.8048780487804879",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "frontiermath_t13",
   "score": 0.887719298245614,
   "source_value_text": "0.887719298245614",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "gpqa_diamond",
   "score": 0.9558080808080808,
   "source_value_text": "0.9558080808080808",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.92929275,
   "source_value_text": "0.92929275",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "simpleqa_verified",
   "score": 0.465,
   "source_value_text": "0.465",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-29"
  },
  {
   "model": "claude-sonnet-5-5",
   "benchmark": "or_tau2_airline",
   "score": 0.775,
   "source_value_text": "0.775",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "chess_puzzles",
   "score": 0.56,
   "source_value_text": "0.56",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "furniture_assembly",
   "score": 0.5833333333333334,
   "source_value_text": "0.5833333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-23"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "mystery_game_puzzles",
   "score": 0.56,
   "source_value_text": "0.56",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "frontiermath_t4",
   "score": 0.9,
   "source_value_text": "0.9",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "frontiermath_t13",
   "score": 0.8982456140350877,
   "source_value_text": "0.8982456140350877",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "gpqa_diamond",
   "score": 0.9431818181818182,
   "source_value_text": "0.9431818181818182",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9208749999999999,
   "source_value_text": "0.9208749999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "simpleqa_verified",
   "score": 0.607,
   "source_value_text": "0.607",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-sol",
   "benchmark": "or_tau2_airline",
   "score": 0.7022223333333333,
   "source_value_text": "0.7022223333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "chess_puzzles",
   "score": 0.59,
   "source_value_text": "0.59",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "furniture_assembly",
   "score": 0.725,
   "source_value_text": "0.725",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "mystery_game_puzzles",
   "score": 0.64,
   "source_value_text": "0.64",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "frontiermath_t4",
   "score": 0.7804878048780488,
   "source_value_text": "0.7804878048780488",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "frontiermath_t13",
   "score": 0.8596491228070176,
   "source_value_text": "0.8596491228070176",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "gpqa_diamond",
   "score": 0.9640151515151515,
   "source_value_text": "0.9640151515151515",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "or_gpqa_diamond",
   "score": 0.964646,
   "source_value_text": "0.964646",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "simpleqa_verified",
   "score": 0.76,
   "source_value_text": "0.76",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-4-argon",
   "benchmark": "or_tau2_airline",
   "score": 0.786667,
   "source_value_text": "0.786667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "chess_puzzles",
   "score": 0.59,
   "source_value_text": "0.59",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "furniture_assembly",
   "score": 0.5666666666666667,
   "source_value_text": "0.5666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "mystery_game_puzzles",
   "score": 0.58,
   "source_value_text": "0.58",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-28"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "frontiermath_t4",
   "score": 0.8292682926829268,
   "source_value_text": "0.8292682926829268",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "frontiermath_t13",
   "score": 0.8912280701754386,
   "source_value_text": "0.8912280701754386",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "gpqa_diamond",
   "score": 0.9349747474747475,
   "source_value_text": "0.9349747474747475",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9234005000000001,
   "source_value_text": "0.9234005000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "simpleqa_verified",
   "score": 0.697,
   "source_value_text": "0.697",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "gpt-5-6-sol",
   "benchmark": "or_tau2_airline",
   "score": 0.73333325,
   "source_value_text": "0.73333325",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "chess_puzzles",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "furniture_assembly",
   "score": 0.7,
   "source_value_text": "0.7",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "mystery_game_puzzles",
   "score": 0.58,
   "source_value_text": "0.58",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "frontiermath_t4",
   "score": 0.8780487804878049,
   "source_value_text": "0.8780487804878049",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "frontiermath_t13",
   "score": 0.9017543859649123,
   "source_value_text": "0.9017543859649123",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8894498333333334,
   "source_value_text": "0.8894498333333334",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "simpleqa_verified",
   "score": 0.708,
   "source_value_text": "0.708",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "claude-fable-5-1",
   "benchmark": "or_tau2_airline",
   "score": 0.7906666,
   "source_value_text": "0.7906666",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "chess_puzzles",
   "score": 0.63,
   "source_value_text": "0.63",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "furniture_assembly",
   "score": 0.44166666666666665,
   "source_value_text": "0.44166666666666665",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.56,
   "source_value_text": "0.56",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-24"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "frontiermath_t4",
   "score": 0.725,
   "source_value_text": "0.725",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "frontiermath_t13",
   "score": 0.8526315789473684,
   "source_value_text": "0.8526315789473684",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-24"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "swe_bench_verified",
   "score": 0.8057851239669421,
   "source_value_text": "0.8057851239669421",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-24"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "gpqa_diamond",
   "score": 0.9400252525252525,
   "source_value_text": "0.9400252525252525",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-24"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9360266666666668,
   "source_value_text": "0.9360266666666668",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "simpleqa_verified",
   "score": 0.63,
   "source_value_text": "0.63",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-5",
   "benchmark": "or_tau2_airline",
   "score": 0.75,
   "source_value_text": "0.75",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "chess_puzzles",
   "score": 0.47,
   "source_value_text": "0.47",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "furniture_assembly",
   "score": 0.35833333333333334,
   "source_value_text": "0.35833333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.52,
   "source_value_text": "0.52",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-31"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "frontiermath_t4",
   "score": 0.902,
   "source_value_text": "0.902",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-09"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "frontiermath_t13",
   "score": 0.8701754385964913,
   "source_value_text": "0.8701754385964913",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-09"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "gpqa_diamond",
   "score": 0.8585858585858586,
   "source_value_text": "0.8585858585858586",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8615317499999998,
   "source_value_text": "0.8615317499999998",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "simpleqa_verified",
   "score": 0.707,
   "source_value_text": "0.707",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "claude-fable-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7957576363636364,
   "source_value_text": "0.7957576363636364",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "chess_puzzles",
   "score": 0.43,
   "source_value_text": "0.43",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "furniture_assembly",
   "score": 0.6083333333333333,
   "source_value_text": "0.6083333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.59,
   "source_value_text": "0.59",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "frontiermath_t4",
   "score": 0.7317073170731707,
   "source_value_text": "0.7317073170731707",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-24"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "frontiermath_t13",
   "score": 0.856140350877193,
   "source_value_text": "0.856140350877193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-24"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "otis_mock_aime",
   "score": 0.9888888888888889,
   "source_value_text": "0.9888888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-24"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "gpqa_diamond",
   "score": 0.9387626262626263,
   "source_value_text": "0.9387626262626263",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-24"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8957629166666667,
   "source_value_text": "0.8957629166666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "simpleqa_verified",
   "score": 0.599,
   "source_value_text": "0.599",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "claude-opus-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7727272727272727,
   "source_value_text": "0.7727272727272727",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "chess_puzzles",
   "score": 0.47,
   "source_value_text": "0.47",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "furniture_assembly",
   "score": 0.5416666666666666,
   "source_value_text": "0.5416666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "mystery_game_puzzles",
   "score": 0.35,
   "source_value_text": "0.35",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-28"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "frontiermath_t4",
   "score": 0.7073170731707317,
   "source_value_text": "0.7073170731707317",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "frontiermath_t13",
   "score": 0.8596491228070176,
   "source_value_text": "0.8596491228070176",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "otis_mock_aime",
   "score": 0.9972222222222222,
   "source_value_text": "0.9972222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "gpqa_diamond",
   "score": 0.9330808080808081,
   "source_value_text": "0.9330808080808081",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "or_gpqa_diamond",
   "score": 0.89267675,
   "source_value_text": "0.89267675",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "simpleqa_verified",
   "score": 0.432,
   "source_value_text": "0.432",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "gpt-5-6-terra",
   "benchmark": "or_tau2_airline",
   "score": 0.715,
   "source_value_text": "0.715",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "chess_puzzles",
   "score": 0.52,
   "source_value_text": "0.52",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "furniture_assembly",
   "score": 0.425,
   "source_value_text": "0.425",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "mystery_game_puzzles",
   "score": 0.36,
   "source_value_text": "0.36",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "frontiermath_t4",
   "score": 0.5609756097560976,
   "source_value_text": "0.5609756097560976",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "frontiermath_t13",
   "score": 0.8,
   "source_value_text": "0.8",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "otis_mock_aime",
   "score": 0.9833333333333333,
   "source_value_text": "0.9833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-07"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "gpqa_diamond",
   "score": 0.9103535353535354,
   "source_value_text": "0.9103535353535354",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-07"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8959834999999999,
   "source_value_text": "0.8959834999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "simpleqa_verified",
   "score": 0.53,
   "source_value_text": "0.53",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "claude-opus-4-8",
   "benchmark": "or_tau2_airline",
   "score": 0.7776190714285715,
   "source_value_text": "0.7776190714285715",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "chess_puzzles",
   "score": 0.66,
   "source_value_text": "0.66",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "furniture_assembly",
   "score": 0.26666666666666666,
   "source_value_text": "0.26666666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.36585365853658536,
   "source_value_text": "0.36585365853658536",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.7157894736842105,
   "source_value_text": "0.7157894736842105",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9722222222222222,
   "source_value_text": "0.9722222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.9482323232323232,
   "source_value_text": "0.9482323232323232",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9391951653543308,
   "source_value_text": "0.9391951653543308",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.692,
   "source_value_text": "0.692",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemini-3-7-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.7831052054794521,
   "source_value_text": "0.7831052054794521",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "chess_puzzles",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "furniture_assembly",
   "score": 0.375,
   "source_value_text": "0.375",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "mystery_game_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-24"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "frontiermath_t4",
   "score": 0.49,
   "source_value_text": "0.49",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "frontiermath_t13",
   "score": 0.7859649122807018,
   "source_value_text": "0.7859649122807018",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "otis_mock_aime",
   "score": 0.9777777777777777,
   "source_value_text": "0.9777777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-15"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "swe_bench_verified",
   "score": 0.768595041322314,
   "source_value_text": "0.768595041322314",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-03-06"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "gpqa_diamond",
   "score": 0.933,
   "source_value_text": "0.933",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-03-06"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "or_gpqa_diamond",
   "score": 0.899551,
   "source_value_text": "0.899551",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "simpleqa_verified",
   "score": 0.451,
   "source_value_text": "0.451",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-4",
   "benchmark": "or_tau2_airline",
   "score": 0.73166675,
   "source_value_text": "0.73166675",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "chess_puzzles",
   "score": 0.43,
   "source_value_text": "0.43",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "furniture_assembly",
   "score": 0.425,
   "source_value_text": "0.425",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "mystery_game_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-28"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "frontiermath_t4",
   "score": 0.6097560975609756,
   "source_value_text": "0.6097560975609756",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "frontiermath_t13",
   "score": 0.8210526315789474,
   "source_value_text": "0.8210526315789474",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "otis_mock_aime",
   "score": 0.9833333333333333,
   "source_value_text": "0.9833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "gpqa_diamond",
   "score": 0.9160353535353535,
   "source_value_text": "0.9160353535353535",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8724745,
   "source_value_text": "0.8724745",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "simpleqa_verified",
   "score": 0.41,
   "source_value_text": "0.41",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "gpt-5-6-luna",
   "benchmark": "or_tau2_airline",
   "score": 0.72333325,
   "source_value_text": "0.72333325",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "chess_puzzles",
   "score": 0.36,
   "source_value_text": "0.36",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "furniture_assembly",
   "score": 0.44166666666666665,
   "source_value_text": "0.44166666666666665",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-28"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "mystery_game_puzzles",
   "score": 0.07,
   "source_value_text": "0.07",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "frontiermath_t4",
   "score": 0.5609756097560976,
   "source_value_text": "0.5609756097560976",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "frontiermath_t13",
   "score": 0.7894736842105263,
   "source_value_text": "0.7894736842105263",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "otis_mock_aime",
   "score": 0.9888888888888889,
   "source_value_text": "0.9888888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "gpqa_diamond",
   "score": 0.9046717171717171,
   "source_value_text": "0.9046717171717171",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8815936666666665,
   "source_value_text": "0.8815936666666665",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "simpleqa_verified",
   "score": 0.414,
   "source_value_text": "0.414",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gpt-6-luna",
   "benchmark": "or_tau2_airline",
   "score": 0.68,
   "source_value_text": "0.68",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "chess_puzzles",
   "score": 0.58,
   "source_value_text": "0.58",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "furniture_assembly",
   "score": 0.31666666666666665,
   "source_value_text": "0.31666666666666665",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.47,
   "source_value_text": "0.47",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.21951219512195122,
   "source_value_text": "0.21951219512195122",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.6842105263157895,
   "source_value_text": "0.6842105263157895",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9888888888888889,
   "source_value_text": "0.9888888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.9539141414141414,
   "source_value_text": "0.9539141414141414",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9482614885496183,
   "source_value_text": "0.9482614885496183",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.697,
   "source_value_text": "0.697",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "gemini-3-8-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.734694,
   "source_value_text": "0.734694",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "chess_puzzles",
   "score": 0.55,
   "source_value_text": "0.55",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "furniture_assembly",
   "score": 0.475,
   "source_value_text": "0.475",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-07"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.3,
   "source_value_text": "0.3",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-07"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "frontiermath_t4",
   "score": 0.4634146341463415,
   "source_value_text": "0.4634146341463415",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-07"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "frontiermath_t13",
   "score": 0.7508771929824561,
   "source_value_text": "0.7508771929824561",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-07"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "otis_mock_aime",
   "score": 0.9888888888888889,
   "source_value_text": "0.9888888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-07"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "gpqa_diamond",
   "score": 0.8964646464646465,
   "source_value_text": "0.8964646464646465",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-09"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8692479999999999,
   "source_value_text": "0.8692479999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "simpleqa_verified",
   "score": 0.238,
   "source_value_text": "0.238",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-09"
  },
  {
   "model": "claude-haiku-5-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7577776666666667,
   "source_value_text": "0.7577776666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "chess_puzzles",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "furniture_assembly",
   "score": 0.3416666666666667,
   "source_value_text": "0.3416666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.43,
   "source_value_text": "0.43",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.2682926829268293,
   "source_value_text": "0.2682926829268293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.6736842105263158,
   "source_value_text": "0.6736842105263158",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9833333333333333,
   "source_value_text": "0.9833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.8983585858585859,
   "source_value_text": "0.8983585858585859",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-09"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8838381666666666,
   "source_value_text": "0.8838381666666666",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.429,
   "source_value_text": "0.429",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "deepseek-v4-1-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.745,
   "source_value_text": "0.745",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "chess_puzzles",
   "score": 0.46,
   "source_value_text": "0.46",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "mystery_game_puzzles",
   "score": 0.43,
   "source_value_text": "0.43",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-19"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "frontiermath_t4",
   "score": 0.2682926829268293,
   "source_value_text": "0.2682926829268293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-19"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "frontiermath_t13",
   "score": 0.6456140350877193,
   "source_value_text": "0.6456140350877193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-19"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "otis_mock_aime",
   "score": 0.9861111111111112,
   "source_value_text": "0.9861111111111112",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-18"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "gpqa_diamond",
   "score": 0.9166666666666666,
   "source_value_text": "0.9166666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-18"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "or_gpqa_diamond",
   "score": 0.892567903763697,
   "source_value_text": "0.892567903763697",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-09-19"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "simpleqa_verified",
   "score": 0.5290581162324649,
   "source_value_text": "0.5290581162324649",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "deepseek-v4-pro-0813",
   "benchmark": "or_tau2_airline",
   "score": 0.767500125,
   "source_value_text": "0.767500125",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "frontiermath_t4",
   "score": 0.3170731707317073,
   "source_value_text": "0.3170731707317073",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "frontiermath_t13",
   "score": 0.6982456140350877,
   "source_value_text": "0.6982456140350877",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "otis_mock_aime",
   "score": 0.9805555555555555,
   "source_value_text": "0.9805555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "gpqa_diamond",
   "score": 0.8857323232323232,
   "source_value_text": "0.8857323232323232",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "or_gpqa_diamond",
   "score": 0.905724,
   "source_value_text": "0.905724",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "simpleqa_verified",
   "score": 0.566,
   "source_value_text": "0.566",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "muse-spark-open",
   "benchmark": "or_tau2_airline",
   "score": 0.78,
   "source_value_text": "0.78",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "kimi-k3",
   "benchmark": "chess_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "kimi-k3",
   "benchmark": "furniture_assembly",
   "score": 0.3416666666666667,
   "source_value_text": "0.3416666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-11"
  },
  {
   "model": "kimi-k3",
   "benchmark": "mystery_game_puzzles",
   "score": 0.26,
   "source_value_text": "0.26",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-29"
  },
  {
   "model": "kimi-k3",
   "benchmark": "frontiermath_t4",
   "score": 0.3902439024390244,
   "source_value_text": "0.3902439024390244",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-17"
  },
  {
   "model": "kimi-k3",
   "benchmark": "frontiermath_t13",
   "score": 0.721830985915493,
   "source_value_text": "0.721830985915493",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-17"
  },
  {
   "model": "kimi-k3",
   "benchmark": "otis_mock_aime",
   "score": 0.9722222222222222,
   "source_value_text": "0.9722222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-16"
  },
  {
   "model": "kimi-k3",
   "benchmark": "gpqa_diamond",
   "score": 0.9311868686868687,
   "source_value_text": "0.9311868686868687",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-16"
  },
  {
   "model": "kimi-k3",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9157357830756013,
   "source_value_text": "0.9157357830756013",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "kimi-k3",
   "benchmark": "simpleqa_verified",
   "score": 0.506,
   "source_value_text": "0.506",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "kimi-k3",
   "benchmark": "or_tau2_airline",
   "score": 0.7347416579476861,
   "source_value_text": "0.7347416579476861",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "chess_puzzles",
   "score": 0.41,
   "source_value_text": "0.41",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "furniture_assembly",
   "score": 0.3333333333333333,
   "source_value_text": "0.3333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "mystery_game_puzzles",
   "score": 0.28,
   "source_value_text": "0.28",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "frontiermath_t4",
   "score": 0.3170731707317073,
   "source_value_text": "0.3170731707317073",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "frontiermath_t13",
   "score": 0.7017543859649122,
   "source_value_text": "0.7017543859649122",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "otis_mock_aime",
   "score": 0.978,
   "source_value_text": "0.978",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-17"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "swe_bench_verified",
   "score": 0.8347107438016529,
   "source_value_text": "0.8347107438016529",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-20"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "gpqa_diamond",
   "score": 0.9015151515151515,
   "source_value_text": "0.9015151515151515",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-17"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "or_gpqa_diamond",
   "score": 0.892375117408907,
   "source_value_text": "0.892375117408907",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "simpleqa_verified",
   "score": 0.517,
   "source_value_text": "0.517",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "claude-opus-4-7",
   "benchmark": "or_tau2_airline",
   "score": 0.7757142142857143,
   "source_value_text": "0.7757142142857143",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "chess_puzzles",
   "score": 0.51,
   "source_value_text": "0.51",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "furniture_assembly",
   "score": 0.26666666666666666,
   "source_value_text": "0.26666666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "mystery_game_puzzles",
   "score": 0.34,
   "source_value_text": "0.34",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-27"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "frontiermath_t4",
   "score": 0.2682926829268293,
   "source_value_text": "0.2682926829268293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "frontiermath_t13",
   "score": 0.5964912280701754,
   "source_value_text": "0.5964912280701754",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "otis_mock_aime",
   "score": 0.956,
   "source_value_text": "0.956",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-20"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "swe_bench_verified",
   "score": 0.756198347107438,
   "source_value_text": "0.756198347107438",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-24"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.9444444444444444,
   "source_value_text": "0.9444444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9483723333333332,
   "source_value_text": "0.9483723333333332",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "simpleqa_verified",
   "score": 0.735,
   "source_value_text": "0.735",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "gemini-3-1-pro",
   "benchmark": "or_tau2_airline",
   "score": 0.7422223333333333,
   "source_value_text": "0.7422223333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "chess_puzzles",
   "score": 0.45,
   "source_value_text": "0.45",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.32,
   "source_value_text": "0.32",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-27"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.2682926829268293,
   "source_value_text": "0.2682926829268293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.6280701754385964,
   "source_value_text": "0.6280701754385964",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9555555555555556,
   "source_value_text": "0.9555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-25"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "swe_bench_verified",
   "score": 0.7933884297520661,
   "source_value_text": "0.7933884297520661",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-01"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.928030303030303,
   "source_value_text": "0.928030303030303",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-22"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9242421666666667,
   "source_value_text": "0.9242421666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.662,
   "source_value_text": "0.662",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemini-3-5-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.7333336666666668,
   "source_value_text": "0.7333336666666668",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "grok-4-6",
   "benchmark": "chess_puzzles",
   "score": 0.33,
   "source_value_text": "0.33",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "grok-4-6",
   "benchmark": "furniture_assembly",
   "score": 0.4,
   "source_value_text": "0.4",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-24"
  },
  {
   "model": "grok-4-6",
   "benchmark": "mystery_game_puzzles",
   "score": 0.34,
   "source_value_text": "0.34",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "grok-4-6",
   "benchmark": "frontiermath_t4",
   "score": 0.3170731707317073,
   "source_value_text": "0.3170731707317073",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "grok-4-6",
   "benchmark": "frontiermath_t13",
   "score": 0.6596491228070176,
   "source_value_text": "0.6596491228070176",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "grok-4-6",
   "benchmark": "otis_mock_aime",
   "score": 0.9916666666666667,
   "source_value_text": "0.9916666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "grok-4-6",
   "benchmark": "gpqa_diamond",
   "score": 0.9400252525252525,
   "source_value_text": "0.9400252525252525",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-12"
  },
  {
   "model": "grok-4-6",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9246375695652174,
   "source_value_text": "0.9246375695652174",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "grok-4-6",
   "benchmark": "simpleqa_verified",
   "score": 0.493,
   "source_value_text": "0.493",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "grok-4-6",
   "benchmark": "or_tau2_airline",
   "score": 0.747619,
   "source_value_text": "0.747619",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "chess_puzzles",
   "score": 0.45,
   "source_value_text": "0.45",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.35,
   "source_value_text": "0.35",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-28"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "frontiermath_t4",
   "score": 0.2926829268292683,
   "source_value_text": "0.2926829268292683",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-30"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "frontiermath_t13",
   "score": 0.656140350877193,
   "source_value_text": "0.656140350877193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-30"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "otis_mock_aime",
   "score": 0.9472222222222222,
   "source_value_text": "0.9472222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-01"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "gpqa_diamond",
   "score": 0.9053030303030303,
   "source_value_text": "0.9053030303030303",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-01"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8576035833333334,
   "source_value_text": "0.8576035833333334",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "simpleqa_verified",
   "score": 0.337,
   "source_value_text": "0.337",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "claude-sonnet-5",
   "benchmark": "or_tau2_airline",
   "score": 0.76,
   "source_value_text": "0.76",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-5-3",
   "benchmark": "chess_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "glm-5-3",
   "benchmark": "mystery_game_puzzles",
   "score": 0.33,
   "source_value_text": "0.33",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "glm-5-3",
   "benchmark": "frontiermath_t4",
   "score": 0.2926829268292683,
   "source_value_text": "0.2926829268292683",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-25"
  },
  {
   "model": "glm-5-3",
   "benchmark": "frontiermath_t13",
   "score": 0.6877192982456141,
   "source_value_text": "0.6877192982456141",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-25"
  },
  {
   "model": "glm-5-3",
   "benchmark": "otis_mock_aime",
   "score": 0.9111111111111111,
   "source_value_text": "0.9111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-24"
  },
  {
   "model": "glm-5-3",
   "benchmark": "gpqa_diamond",
   "score": 0.9090909090909091,
   "source_value_text": "0.9090909090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-24"
  },
  {
   "model": "glm-5-3",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8639263412162163,
   "source_value_text": "0.8639263412162163",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-5-3",
   "benchmark": "simpleqa_verified",
   "score": 0.41,
   "source_value_text": "0.41",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "glm-5-3",
   "benchmark": "or_tau2_airline",
   "score": 0.78083325,
   "source_value_text": "0.78083325",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "chess_puzzles",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "furniture_assembly",
   "score": 0.38333333333333336,
   "source_value_text": "0.38333333333333336",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "mystery_game_puzzles",
   "score": 0.23,
   "source_value_text": "0.23",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "frontiermath_t4",
   "score": 0.317,
   "source_value_text": "0.317",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "frontiermath_t13",
   "score": 0.674,
   "source_value_text": "0.674",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "otis_mock_aime",
   "score": 0.9611111111111111,
   "source_value_text": "0.9611111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-12-11"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "swe_bench_verified",
   "score": 0.737603305785124,
   "source_value_text": "0.737603305785124",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-12"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "gpqa_diamond",
   "score": 0.914,
   "source_value_text": "0.914",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-12-13"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "or_gpqa_diamond",
   "score": 0.878788,
   "source_value_text": "0.878788",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "simpleqa_verified",
   "score": 0.371,
   "source_value_text": "0.371",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-2",
   "benchmark": "bfcl_overall",
   "score": 0.5587,
   "source_value_text": "55.87%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-5-2",
   "benchmark": "or_tau2_airline",
   "score": 0.726667,
   "source_value_text": "0.726667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "chess_puzzles",
   "score": 0.26,
   "source_value_text": "0.26",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "mystery_game_puzzles",
   "score": 0.32,
   "source_value_text": "0.32",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-28"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "frontiermath_t4",
   "score": 0.34146341463414637,
   "source_value_text": "0.34146341463414637",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "frontiermath_t13",
   "score": 0.6456140350877193,
   "source_value_text": "0.6456140350877193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "otis_mock_aime",
   "score": 0.9555555555555556,
   "source_value_text": "0.9555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "swe_bench_verified",
   "score": 0.7727272727272727,
   "source_value_text": "0.7727272727272727",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-18"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "gpqa_diamond",
   "score": 0.9090909090909091,
   "source_value_text": "0.9090909090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9090909999999999,
   "source_value_text": "0.9090909999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-09-27"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "simpleqa_verified",
   "score": 0.5576730190571715,
   "source_value_text": "0.5576730190571715",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-7-max",
   "benchmark": "or_tau2_airline",
   "score": 0.72,
   "source_value_text": "0.72",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-09-27"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "chess_puzzles",
   "score": 0.42,
   "source_value_text": "0.42",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "furniture_assembly",
   "score": 0.23333333333333334,
   "source_value_text": "0.23333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.3,
   "source_value_text": "0.3",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.21951219512195122,
   "source_value_text": "0.21951219512195122",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.5894736842105263,
   "source_value_text": "0.5894736842105263",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9416666666666667,
   "source_value_text": "0.9416666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.9412878787878788,
   "source_value_text": "0.9412878787878788",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.930415,
   "source_value_text": "0.930415",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.662,
   "source_value_text": "0.662",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemini-3-6-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.72,
   "source_value_text": "0.72",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "chess_puzzles",
   "score": 0.32,
   "source_value_text": "0.32",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "mystery_game_puzzles",
   "score": 0.34,
   "source_value_text": "0.34",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "frontiermath_t4",
   "score": 0.24390243902439024,
   "source_value_text": "0.24390243902439024",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "frontiermath_t13",
   "score": 0.5754385964912281,
   "source_value_text": "0.5754385964912281",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "otis_mock_aime",
   "score": 0.9444444444444444,
   "source_value_text": "0.9444444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "gpqa_diamond",
   "score": 0.9103535353535354,
   "source_value_text": "0.9103535353535354",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8996633999999999,
   "source_value_text": "0.8996633999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-08"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "simpleqa_verified",
   "score": 0.33633633633633636,
   "source_value_text": "0.33633633633633636",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "deepseek-v4-flash-0731",
   "benchmark": "or_tau2_airline",
   "score": 0.7766665,
   "source_value_text": "0.7766665",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-08"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "chess_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "furniture_assembly",
   "score": 0.2833333333333333,
   "source_value_text": "0.2833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "mystery_game_puzzles",
   "score": 0.25,
   "source_value_text": "0.25",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "frontiermath_t4",
   "score": 0.2682926829268293,
   "source_value_text": "0.2682926829268293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "frontiermath_t13",
   "score": 0.6596491228070176,
   "source_value_text": "0.6596491228070176",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "otis_mock_aime",
   "score": 0.9444444444444444,
   "source_value_text": "0.9444444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-06"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "swe_bench_verified",
   "score": 0.7871900826446281,
   "source_value_text": "0.7871900826446281",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-18"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "gpqa_diamond",
   "score": 0.9053030303030303,
   "source_value_text": "0.9053030303030303",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-06"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "or_gpqa_diamond",
   "score": 0.856643153846154,
   "source_value_text": "0.856643153846154",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "simpleqa_verified",
   "score": 0.47,
   "source_value_text": "0.47",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "claude-opus-4-6",
   "benchmark": "or_tau2_airline",
   "score": 0.7638095,
   "source_value_text": "0.7638095",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "chess_puzzles",
   "score": 0.39,
   "source_value_text": "0.39",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.26,
   "source_value_text": "0.26",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.17073170731707318,
   "source_value_text": "0.17073170731707318",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.512280701754386,
   "source_value_text": "0.512280701754386",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9555555555555556,
   "source_value_text": "0.9555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "swe_bench_verified",
   "score": 0.7541322314049587,
   "source_value_text": "0.7541322314049587",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-18"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.8939393939393939,
   "source_value_text": "0.8939393939393939",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8855220000000001,
   "source_value_text": "0.8855220000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.668,
   "source_value_text": "0.668",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemini-3-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.7533333333333333,
   "source_value_text": "0.7533333333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "chess_puzzles",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "furniture_assembly",
   "score": 0.21666666666666667,
   "source_value_text": "0.21666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-11"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "mystery_game_puzzles",
   "score": 0.18,
   "source_value_text": "0.18",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-17"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "frontiermath_t4",
   "score": 0.2564102564102564,
   "source_value_text": "0.2564102564102564",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "frontiermath_t13",
   "score": 0.5719298245614035,
   "source_value_text": "0.5719298245614035",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "otis_mock_aime",
   "score": 0.9611111111111111,
   "source_value_text": "0.9611111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-02"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "swe_bench_verified",
   "score": 0.7665289256198347,
   "source_value_text": "0.7665289256198347",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-08"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "gpqa_diamond",
   "score": 0.9078282828282829,
   "source_value_text": "0.9078282828282829",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-01"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8799901428571427,
   "source_value_text": "0.8799901428571427",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "simpleqa_verified",
   "score": 0.349,
   "source_value_text": "0.349",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "kimi-k2-6",
   "benchmark": "or_tau2_airline",
   "score": 0.7671130223214285,
   "source_value_text": "0.7671130223214285",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-5",
   "benchmark": "chess_puzzles",
   "score": 0.36,
   "source_value_text": "0.36",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.23,
   "source_value_text": "0.23",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "gpt-5",
   "benchmark": "frontiermath_t4",
   "score": 0.21951219512195122,
   "source_value_text": "0.21951219512195122",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gpt-5",
   "benchmark": "frontiermath_t13",
   "score": 0.5543859649122806,
   "source_value_text": "0.5543859649122806",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-10"
  },
  {
   "model": "gpt-5",
   "benchmark": "math_level_5",
   "score": 0.9813066465256798,
   "source_value_text": "0.9813066465256798",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-29"
  },
  {
   "model": "gpt-5",
   "benchmark": "otis_mock_aime",
   "score": 0.9138888888888889,
   "source_value_text": "0.9138888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-29"
  },
  {
   "model": "gpt-5",
   "benchmark": "aider_polyglot",
   "score": 0.88,
   "source_value_text": "88.0",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-08-23"
  },
  {
   "model": "gpt-5",
   "benchmark": "swe_bench_verified",
   "score": 0.7355371900826446,
   "source_value_text": "0.7355371900826446",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-06"
  },
  {
   "model": "gpt-5",
   "benchmark": "gpqa_diamond",
   "score": 0.8617424242424242,
   "source_value_text": "0.8617424242424242",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-29"
  },
  {
   "model": "gpt-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8501685000000001,
   "source_value_text": "0.8501685000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gpt-5",
   "benchmark": "simpleqa_verified",
   "score": 0.501,
   "source_value_text": "0.501",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7066666666666667,
   "source_value_text": "0.7066666666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "glm-5-2",
   "benchmark": "chess_puzzles",
   "score": 0.22,
   "source_value_text": "0.22",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "glm-5-2",
   "benchmark": "mystery_game_puzzles",
   "score": 0.19,
   "source_value_text": "0.19",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-5-2",
   "benchmark": "frontiermath_t4",
   "score": 0.2926829268292683,
   "source_value_text": "0.2926829268292683",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-19"
  },
  {
   "model": "glm-5-2",
   "benchmark": "frontiermath_t13",
   "score": 0.592057761732852,
   "source_value_text": "0.592057761732852",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-19"
  },
  {
   "model": "glm-5-2",
   "benchmark": "otis_mock_aime",
   "score": 0.8638888888888889,
   "source_value_text": "0.8638888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-25"
  },
  {
   "model": "glm-5-2",
   "benchmark": "swe_bench_verified",
   "score": 0.787,
   "source_value_text": "0.787",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-25"
  },
  {
   "model": "glm-5-2",
   "benchmark": "gpqa_diamond",
   "score": 0.9185606060606061,
   "source_value_text": "0.9185606060606061",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-24"
  },
  {
   "model": "glm-5-2",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8334862727272728,
   "source_value_text": "0.8334862727272728",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-5-2",
   "benchmark": "simpleqa_verified",
   "score": 0.342,
   "source_value_text": "0.342",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-5-2",
   "benchmark": "or_tau2_airline",
   "score": 0.7169697272727272,
   "source_value_text": "0.7169697272727272",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "chess_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.08,
   "source_value_text": "0.08",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "frontiermath_t4",
   "score": 0.17073170731707318,
   "source_value_text": "0.17073170731707318",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.5578947368421052,
   "source_value_text": "0.5578947368421052",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.9388888888888889,
   "source_value_text": "0.9388888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-26"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.9015151515151515,
   "source_value_text": "0.9015151515151515",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-26"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9090909999999999,
   "source_value_text": "0.9090909999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "glm-5-3-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.762499875,
   "source_value_text": "0.762499875",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "chess_puzzles",
   "score": 0.42,
   "source_value_text": "0.42",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "mystery_game_puzzles",
   "score": 0.19,
   "source_value_text": "0.19",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "otis_mock_aime",
   "score": 0.8861111111111111,
   "source_value_text": "0.8861111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-13"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "swe_bench_verified",
   "score": 0.6797520661157025,
   "source_value_text": "0.6797520661157025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-18"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "gpqa_diamond",
   "score": 0.8762626262626263,
   "source_value_text": "0.8762626262626263",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-13"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8692479999999999,
   "source_value_text": "0.8692479999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "simpleqa_verified",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-1",
   "benchmark": "or_tau2_airline",
   "score": 0.7583332500000001,
   "source_value_text": "0.7583332500000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "kimi-k2-5",
   "benchmark": "otis_mock_aime",
   "score": 0.922,
   "source_value_text": "0.922",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-02"
  },
  {
   "model": "kimi-k2-5",
   "benchmark": "swe_bench_verified",
   "score": 0.737603305785124,
   "source_value_text": "0.737603305785124",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-17"
  },
  {
   "model": "kimi-k2-5",
   "benchmark": "gpqa_diamond",
   "score": 0.876,
   "source_value_text": "0.876",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-02"
  },
  {
   "model": "kimi-k2-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8592592,
   "source_value_text": "0.8592592",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "kimi-k2-5",
   "benchmark": "simpleqa_verified",
   "score": 0.343,
   "source_value_text": "0.343",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "kimi-k2-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7277775,
   "source_value_text": "0.7277775",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "chess_puzzles",
   "score": 0.14,
   "source_value_text": "0.14",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "frontiermath_t4",
   "score": 0.12195121951219512,
   "source_value_text": "0.12195121951219512",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "frontiermath_t13",
   "score": 0.5403508771929825,
   "source_value_text": "0.5403508771929825",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "otis_mock_aime",
   "score": 0.9555555555555556,
   "source_value_text": "0.9555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "gpqa_diamond",
   "score": 0.8787878787878788,
   "source_value_text": "0.8787878787878788",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8847767229341582,
   "source_value_text": "0.8847767229341582",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "simpleqa_verified",
   "score": 0.365,
   "source_value_text": "0.365",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "kimi-k2-7-code",
   "benchmark": "or_tau2_airline",
   "score": 0.7383231627502177,
   "source_value_text": "0.7383231627502177",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "inkling-small",
   "benchmark": "mystery_game_puzzles",
   "score": 0.06,
   "source_value_text": "0.06",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-15"
  },
  {
   "model": "inkling-small",
   "benchmark": "frontiermath_t4",
   "score": 0.17073170731707318,
   "source_value_text": "0.17073170731707318",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "inkling-small",
   "benchmark": "frontiermath_t13",
   "score": 0.4631578947368421,
   "source_value_text": "0.4631578947368421",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "inkling-small",
   "benchmark": "otis_mock_aime",
   "score": 0.9,
   "source_value_text": "0.9",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "inkling-small",
   "benchmark": "gpqa_diamond",
   "score": 0.88510101010101,
   "source_value_text": "0.88510101010101",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-14"
  },
  {
   "model": "inkling-small",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8703701666666666,
   "source_value_text": "0.8703701666666666",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-09-19"
  },
  {
   "model": "inkling-small",
   "benchmark": "simpleqa_verified",
   "score": 0.191,
   "source_value_text": "0.191",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "inkling-small",
   "benchmark": "or_tau2_airline",
   "score": 0.6076188571428572,
   "source_value_text": "0.6076188571428572",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "chess_puzzles",
   "score": 0.45,
   "source_value_text": "0.45",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "mystery_game_puzzles",
   "score": 0.09,
   "source_value_text": "0.09",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "frontiermath_t4",
   "score": 0.12195121951219512,
   "source_value_text": "0.12195121951219512",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "frontiermath_t13",
   "score": 0.44912280701754387,
   "source_value_text": "0.44912280701754387",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "otis_mock_aime",
   "score": 0.8778,
   "source_value_text": "0.8778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-14"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "gpqa_diamond",
   "score": 0.7847,
   "source_value_text": "0.7847",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-14"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "or_gpqa_diamond",
   "score": 0.77567325,
   "source_value_text": "0.77567325",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "simpleqa_verified",
   "score": 0.117,
   "source_value_text": "0.117",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-4-nano",
   "benchmark": "or_tau2_airline",
   "score": 0.65166675,
   "source_value_text": "0.65166675",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "chess_puzzles",
   "score": 0.23,
   "source_value_text": "0.23",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "mystery_game_puzzles",
   "score": 0.11,
   "source_value_text": "0.11",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "frontiermath_t4",
   "score": 0.0975609756097561,
   "source_value_text": "0.0975609756097561",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "frontiermath_t13",
   "score": 0.512280701754386,
   "source_value_text": "0.512280701754386",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.8888888888888888,
   "source_value_text": "0.8888888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.8686868686868687,
   "source_value_text": "0.8686868686868687",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "or_gpqa_diamond",
   "score": 0.84048825,
   "source_value_text": "0.84048825",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "simpleqa_verified",
   "score": 0.294,
   "source_value_text": "0.294",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-4-mini",
   "benchmark": "or_tau2_airline",
   "score": 0.6216667499999999,
   "source_value_text": "0.6216667499999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "chess_puzzles",
   "score": 0.19,
   "source_value_text": "0.19",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "mystery_game_puzzles",
   "score": 0.17,
   "source_value_text": "0.17",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "frontiermath_t4",
   "score": 0.024390243902439025,
   "source_value_text": "0.024390243902439025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "frontiermath_t13",
   "score": 0.45263157894736844,
   "source_value_text": "0.45263157894736844",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "otis_mock_aime",
   "score": 0.9666666666666667,
   "source_value_text": "0.9666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "swe_bench_verified",
   "score": 0.7763975155279503,
   "source_value_text": "0.7763975155279503",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-18"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.9090909090909091,
   "source_value_text": "0.9090909090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "or_gpqa_diamond",
   "score": 0.853535,
   "source_value_text": "0.853535",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-08"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "simpleqa_verified",
   "score": 0.46993987975951906,
   "source_value_text": "0.46993987975951906",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "deepseek-v4-pro",
   "benchmark": "or_tau2_airline",
   "score": 0.788889,
   "source_value_text": "0.788889",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-08"
  },
  {
   "model": "o3",
   "benchmark": "chess_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "o3",
   "benchmark": "mystery_game_puzzles",
   "score": 0.29,
   "source_value_text": "0.29",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "o3",
   "benchmark": "frontiermath_t13",
   "score": 0.3333333333333333,
   "source_value_text": "0.3333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "o3",
   "benchmark": "math_level_5",
   "score": 0.9777190332326284,
   "source_value_text": "0.9777190332326284",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-16"
  },
  {
   "model": "o3",
   "benchmark": "otis_mock_aime",
   "score": 0.8444444444444444,
   "source_value_text": "0.8444444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "o3",
   "benchmark": "aider_polyglot",
   "score": 0.813,
   "source_value_text": "81.3",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-06-25"
  },
  {
   "model": "o3",
   "benchmark": "swe_bench_verified",
   "score": 0.6231884057971014,
   "source_value_text": "0.6231884057971014",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-12"
  },
  {
   "model": "o3",
   "benchmark": "gpqa_diamond",
   "score": 0.8181818181818182,
   "source_value_text": "0.8181818181818182",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-16"
  },
  {
   "model": "o3",
   "benchmark": "simpleqa_verified",
   "score": 0.494,
   "source_value_text": "0.494",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "o3",
   "benchmark": "bfcl_overall",
   "score": 0.6305,
   "source_value_text": "63.05%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "glm-5-1",
   "benchmark": "chess_puzzles",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "glm-5-1",
   "benchmark": "frontiermath_t13",
   "score": 0.3684210526315789,
   "source_value_text": "0.3684210526315789",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "glm-5-1",
   "benchmark": "otis_mock_aime",
   "score": 0.9333333333333333,
   "source_value_text": "0.9333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "glm-5-1",
   "benchmark": "swe_bench_verified",
   "score": 0.7417,
   "source_value_text": "0.7417",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-15"
  },
  {
   "model": "glm-5-1",
   "benchmark": "gpqa_diamond",
   "score": 0.898989898989899,
   "source_value_text": "0.898989898989899",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "glm-5-1",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8115756996197718,
   "source_value_text": "0.8115756996197718",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-5-1",
   "benchmark": "simpleqa_verified",
   "score": 0.34,
   "source_value_text": "0.34",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-5-1",
   "benchmark": "or_tau2_airline",
   "score": 0.7433332500000001,
   "source_value_text": "0.7433332500000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "chess_puzzles",
   "score": 0.24,
   "source_value_text": "0.24",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "mystery_game_puzzles",
   "score": 0.1,
   "source_value_text": "0.1",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "frontiermath_t4",
   "score": 0.12195121951219512,
   "source_value_text": "0.12195121951219512",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "frontiermath_t13",
   "score": 0.4666666666666667,
   "source_value_text": "0.4666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "math_level_5",
   "score": 0.9784743202416919,
   "source_value_text": "0.9784743202416919",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-30"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.8666666666666667,
   "source_value_text": "0.8666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-30"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "swe_bench_verified",
   "score": 0.6466942148760331,
   "source_value_text": "0.6466942148760331",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-01"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.75,
   "source_value_text": "0.75",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-30"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8007853333333332,
   "source_value_text": "0.8007853333333332",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "simpleqa_verified",
   "score": 0.216,
   "source_value_text": "0.216",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "bfcl_overall",
   "score": 0.5546,
   "source_value_text": "55.46%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-5-mini",
   "benchmark": "or_tau2_airline",
   "score": 0.712121101010101,
   "source_value_text": "0.712121101010101",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "qwen3-6-27b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.07,
   "source_value_text": "0.07",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-6-27b",
   "benchmark": "frontiermath_t13",
   "score": 0.3508771929824561,
   "source_value_text": "0.3508771929824561",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-6-27b",
   "benchmark": "otis_mock_aime",
   "score": 0.9111111111111111,
   "source_value_text": "0.9111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-6-27b",
   "benchmark": "gpqa_diamond",
   "score": 0.8585858585858586,
   "source_value_text": "0.8585858585858586",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-6-27b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.801499818181818,
   "source_value_text": "0.801499818181818",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-6-27b",
   "benchmark": "or_tau2_airline",
   "score": 0.7627176144578314,
   "source_value_text": "0.7627176144578314",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-7-plus",
   "benchmark": "mystery_game_puzzles",
   "score": 0.17,
   "source_value_text": "0.17",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-7-plus",
   "benchmark": "frontiermath_t13",
   "score": 0.34385964912280703,
   "source_value_text": "0.34385964912280703",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "qwen3-7-plus",
   "benchmark": "otis_mock_aime",
   "score": 0.9333333333333333,
   "source_value_text": "0.9333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-7-plus",
   "benchmark": "gpqa_diamond",
   "score": 0.8787878787878788,
   "source_value_text": "0.8787878787878788",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-7-plus",
   "benchmark": "or_gpqa_diamond",
   "score": 0.873737,
   "source_value_text": "0.873737",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "qwen3-7-plus",
   "benchmark": "or_tau2_airline",
   "score": 0.726667,
   "source_value_text": "0.726667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "chess_puzzles",
   "score": 0.14,
   "source_value_text": "0.14",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "furniture_assembly",
   "score": 0.2833333333333333,
   "source_value_text": "0.2833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.22,
   "source_value_text": "0.22",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "frontiermath_t4",
   "score": 0.04878048780487805,
   "source_value_text": "0.04878048780487805",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "frontiermath_t13",
   "score": 0.34385964912280703,
   "source_value_text": "0.34385964912280703",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "otis_mock_aime",
   "score": 0.8611111111111112,
   "source_value_text": "0.8611111111111112",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-24"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "swe_bench_verified",
   "score": 0.7665289256198347,
   "source_value_text": "0.7665289256198347",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-05"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "gpqa_diamond",
   "score": 0.860479797979798,
   "source_value_text": "0.860479797979798",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-24"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8610461538461541,
   "source_value_text": "0.8610461538461541",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "simpleqa_verified",
   "score": 0.457,
   "source_value_text": "0.457",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "bfcl_overall",
   "score": 0.7746999999999999,
   "source_value_text": "77.47%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "claude-opus-4-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7700000714285714,
   "source_value_text": "0.7700000714285714",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "inkling",
   "benchmark": "chess_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "inkling",
   "benchmark": "frontiermath_t4",
   "score": 0.04878048780487805,
   "source_value_text": "0.04878048780487805",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "inkling",
   "benchmark": "frontiermath_t13",
   "score": 0.3333333333333333,
   "source_value_text": "0.3333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "inkling",
   "benchmark": "otis_mock_aime",
   "score": 0.8888888888888888,
   "source_value_text": "0.8888888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "inkling",
   "benchmark": "gpqa_diamond",
   "score": 0.8825757575757576,
   "source_value_text": "0.8825757575757576",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "inkling",
   "benchmark": "or_gpqa_diamond",
   "score": 0.818035862576065,
   "source_value_text": "0.818035862576065",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-09-19"
  },
  {
   "model": "inkling",
   "benchmark": "simpleqa_verified",
   "score": 0.403,
   "source_value_text": "0.403",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "inkling",
   "benchmark": "or_tau2_airline",
   "score": 0.6755556666666667,
   "source_value_text": "0.6755556666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "qwen3-5-397b-a17b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.18,
   "source_value_text": "0.18",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-5-397b-a17b",
   "benchmark": "frontiermath_t13",
   "score": 0.312280701754386,
   "source_value_text": "0.312280701754386",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-5-397b-a17b",
   "benchmark": "otis_mock_aime",
   "score": 0.8888888888888888,
   "source_value_text": "0.8888888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-5-397b-a17b",
   "benchmark": "gpqa_diamond",
   "score": 0.8636363636363636,
   "source_value_text": "0.8636363636363636",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-5-397b-a17b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8728152090966519,
   "source_value_text": "0.8728152090966519",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-5-397b-a17b",
   "benchmark": "or_tau2_airline",
   "score": 0.7811111111111111,
   "source_value_text": "0.7811111111111111",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "chess_puzzles",
   "score": 0.23,
   "source_value_text": "0.23",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "mystery_game_puzzles",
   "score": 0.16,
   "source_value_text": "0.16",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "otis_mock_aime",
   "score": 0.857936507936508,
   "source_value_text": "0.857936507936508",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-20"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "swe_bench_verified",
   "score": 0.7520661157024794,
   "source_value_text": "0.7520661157024794",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-21"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "gpqa_diamond",
   "score": 0.8737373737373737,
   "source_value_text": "0.8737373737373737",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-20"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6459467692307692,
   "source_value_text": "0.6459467692307692",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "simpleqa_verified",
   "score": 0.355,
   "source_value_text": "0.355",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "claude-sonnet-4-6",
   "benchmark": "or_tau2_airline",
   "score": 0.7490475714285715,
   "source_value_text": "0.7490475714285715",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-4-7",
   "benchmark": "chess_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "glm-4-7",
   "benchmark": "otis_mock_aime",
   "score": 0.8333333333333334,
   "source_value_text": "0.8333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-01-29"
  },
  {
   "model": "glm-4-7",
   "benchmark": "gpqa_diamond",
   "score": 0.8333333333333334,
   "source_value_text": "0.8333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-01-29"
  },
  {
   "model": "glm-4-7",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8364136921606119,
   "source_value_text": "0.8364136921606119",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-4-7",
   "benchmark": "simpleqa_verified",
   "score": 0.322,
   "source_value_text": "0.322",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-4-7",
   "benchmark": "or_tau2_airline",
   "score": 0.7037037777777777,
   "source_value_text": "0.7037037777777777",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "o4-mini",
   "benchmark": "chess_puzzles",
   "score": 0.22,
   "source_value_text": "0.22",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "o4-mini",
   "benchmark": "mystery_game_puzzles",
   "score": 0.05,
   "source_value_text": "0.05",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "o4-mini",
   "benchmark": "frontiermath_t4",
   "score": 0.04878048780487805,
   "source_value_text": "0.04878048780487805",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "o4-mini",
   "benchmark": "frontiermath_t13",
   "score": 0.36140350877192984,
   "source_value_text": "0.36140350877192984",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "o4-mini",
   "benchmark": "math_level_5",
   "score": 0.978285498489426,
   "source_value_text": "0.978285498489426",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-16"
  },
  {
   "model": "o4-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.8166666666666667,
   "source_value_text": "0.8166666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-16"
  },
  {
   "model": "o4-mini",
   "benchmark": "aider_polyglot",
   "score": 0.72,
   "source_value_text": "72.0",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-16"
  },
  {
   "model": "o4-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.7960858585858586,
   "source_value_text": "0.7960858585858586",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-16"
  },
  {
   "model": "o4-mini",
   "benchmark": "simpleqa_verified",
   "score": 0.196,
   "source_value_text": "0.196",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "o4-mini",
   "benchmark": "bfcl_overall",
   "score": 0.5324,
   "source_value_text": "53.24%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-235b-a22b-thinking-jul-2025",
   "benchmark": "mystery_game_puzzles",
   "score": 0.09,
   "source_value_text": "0.09",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-235b-a22b-thinking-jul-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.8666666666666667,
   "source_value_text": "0.8666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-12-10"
  },
  {
   "model": "qwen3-235b-a22b-thinking-jul-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.8005050505050505,
   "source_value_text": "0.8005050505050505",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-12-11"
  },
  {
   "model": "qwen3-235b-a22b-thinking-jul-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7950335,
   "source_value_text": "0.7950335",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-22"
  },
  {
   "model": "qwen3-235b-a22b-thinking-jul-2025",
   "benchmark": "simpleqa_verified",
   "score": 0.4044044044044044,
   "source_value_text": "0.4044044044044044",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-235b-a22b-thinking-jul-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.6266666666666667,
   "source_value_text": "0.6266666666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-22"
  },
  {
   "model": "qwen-3-6-35b-a3b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.22,
   "source_value_text": "0.22",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen-3-6-35b-a3b",
   "benchmark": "frontiermath_t13",
   "score": 0.20350877192982456,
   "source_value_text": "0.20350877192982456",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen-3-6-35b-a3b",
   "benchmark": "otis_mock_aime",
   "score": 0.8666666666666667,
   "source_value_text": "0.8666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-35b-a3b",
   "benchmark": "gpqa_diamond",
   "score": 0.8484848484848485,
   "source_value_text": "0.8484848484848485",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-35b-a3b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8272847449686571,
   "source_value_text": "0.8272847449686571",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen-3-6-35b-a3b",
   "benchmark": "or_tau2_airline",
   "score": 0.744993393858478,
   "source_value_text": "0.744993393858478",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "chess_puzzles",
   "score": 0.1,
   "source_value_text": "0.1",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "mystery_game_puzzles",
   "score": 0.17,
   "source_value_text": "0.17",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "frontiermath_t4",
   "score": 0.024390243902439025,
   "source_value_text": "0.024390243902439025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "frontiermath_t13",
   "score": 0.23859649122807017,
   "source_value_text": "0.23859649122807017",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "math_level_5",
   "score": 0.9773413897280967,
   "source_value_text": "0.9773413897280967",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-21"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "otis_mock_aime",
   "score": 0.7777777777777778,
   "source_value_text": "0.7777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-28"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "swe_bench_verified",
   "score": 0.7128099173553719,
   "source_value_text": "0.7128099173553719",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-05"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "gpqa_diamond",
   "score": 0.8232323232323232,
   "source_value_text": "0.8232323232323232",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-28"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8111110666666667,
   "source_value_text": "0.8111110666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "simpleqa_verified",
   "score": 0.307,
   "source_value_text": "0.307",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "bfcl_overall",
   "score": 0.7323999999999999,
   "source_value_text": "73.24%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "claude-sonnet-4-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7528572857142857,
   "source_value_text": "0.7528572857142857",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "chess_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "mystery_game_puzzles",
   "score": 0.19,
   "source_value_text": "0.19",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "frontiermath_t4",
   "score": 0.0,
   "source_value_text": "0.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "frontiermath_t13",
   "score": 0.2596491228070175,
   "source_value_text": "0.2596491228070175",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "otis_mock_aime",
   "score": 0.7111111111111111,
   "source_value_text": "0.7111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "gpqa_diamond",
   "score": 0.8333333333333334,
   "source_value_text": "0.8333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8402775531250001,
   "source_value_text": "0.8402775531250001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-3-5-flash-lite",
   "benchmark": "or_tau2_airline",
   "score": 0.7349997500000001,
   "source_value_text": "0.7349997500000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "frontiermath_t4",
   "score": 0.0,
   "source_value_text": "0.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "frontiermath_t13",
   "score": 0.24561403508771928,
   "source_value_text": "0.24561403508771928",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.8472222222222222,
   "source_value_text": "0.8472222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-16"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "aider_polyglot",
   "score": 0.831,
   "source_value_text": "83.1",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-06-06"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "swe_bench_verified",
   "score": 0.5755693581780539,
   "source_value_text": "0.5755693581780539",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-13"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.8529040404040404,
   "source_value_text": "0.8529040404040404",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-16"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.806210222222222,
   "source_value_text": "0.806210222222222",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-2-5-pro-jun-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.5622223333333333,
   "source_value_text": "0.5622223333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "minimax-m3",
   "benchmark": "chess_puzzles",
   "score": 0.13,
   "source_value_text": "0.13",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "minimax-m3",
   "benchmark": "mystery_game_puzzles",
   "score": 0.08,
   "source_value_text": "0.08",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "minimax-m3",
   "benchmark": "otis_mock_aime",
   "score": 0.7111111111111111,
   "source_value_text": "0.7111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "minimax-m3",
   "benchmark": "gpqa_diamond",
   "score": 0.9090909090909091,
   "source_value_text": "0.9090909090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "minimax-m3",
   "benchmark": "or_gpqa_diamond",
   "score": 0.9151233333333333,
   "source_value_text": "0.9151233333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "minimax-m3",
   "benchmark": "or_tau2_airline",
   "score": 0.7388889166666667,
   "source_value_text": "0.7388889166666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-oss-120b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.02,
   "source_value_text": "0.02",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-oss-120b",
   "benchmark": "otis_mock_aime",
   "score": 0.8888888888888888,
   "source_value_text": "0.8888888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-12-11"
  },
  {
   "model": "gpt-oss-120b",
   "benchmark": "aider_polyglot",
   "score": 0.418,
   "source_value_text": "41.8",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-08-06"
  },
  {
   "model": "gpt-oss-120b",
   "benchmark": "gpqa_diamond",
   "score": 0.7575757575757576,
   "source_value_text": "0.7575757575757576",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-12-11"
  },
  {
   "model": "gpt-oss-120b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7578456145038168,
   "source_value_text": "0.7578456145038168",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-oss-120b",
   "benchmark": "or_tau2_airline",
   "score": 0.633333,
   "source_value_text": "0.633333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "chess_puzzles",
   "score": 0.22,
   "source_value_text": "0.22",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "mystery_game_puzzles",
   "score": 0.09,
   "source_value_text": "0.09",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "frontiermath_t4",
   "score": 0.024390243902439025,
   "source_value_text": "0.024390243902439025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "frontiermath_t13",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "math_level_5",
   "score": 0.952416918429003,
   "source_value_text": "0.952416918429003",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-08-20"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "otis_mock_aime",
   "score": 0.8111111111111111,
   "source_value_text": "0.8111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-31"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "gpqa_diamond",
   "score": 0.6944444444444444,
   "source_value_text": "0.6944444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-30"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6843435000000001,
   "source_value_text": "0.6843435000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "simpleqa_verified",
   "score": 0.117,
   "source_value_text": "0.117",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "bfcl_overall",
   "score": 0.5145000000000001,
   "source_value_text": "51.45%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-5-nano",
   "benchmark": "or_tau2_airline",
   "score": 0.4511785656565656,
   "source_value_text": "0.4511785656565656",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "o3-mini",
   "benchmark": "chess_puzzles",
   "score": 0.15,
   "source_value_text": "0.15",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "o3-mini",
   "benchmark": "mystery_game_puzzles",
   "score": 0.07,
   "source_value_text": "0.07",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "o3-mini",
   "benchmark": "frontiermath_t4",
   "score": 0.0,
   "source_value_text": "0.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "o3-mini",
   "benchmark": "frontiermath_t13",
   "score": 0.18596491228070175,
   "source_value_text": "0.18596491228070175",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "o3-mini",
   "benchmark": "math_level_5",
   "score": 0.9648791540785498,
   "source_value_text": "0.9648791540785498",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-13"
  },
  {
   "model": "o3-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.7694444444444445,
   "source_value_text": "0.7694444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-27"
  },
  {
   "model": "o3-mini",
   "benchmark": "aider_polyglot",
   "score": 0.5379999999999999,
   "source_value_text": "53.8",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-01-31"
  },
  {
   "model": "o3-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.7702020202020202,
   "source_value_text": "0.7702020202020202",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-13"
  },
  {
   "model": "o3-mini",
   "benchmark": "simpleqa_verified",
   "score": 0.153,
   "source_value_text": "0.153",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "chess_puzzles",
   "score": 0.08,
   "source_value_text": "0.08",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "math_level_5",
   "score": 0.9636087945413192,
   "source_value_text": "0.9636087945413192",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-22"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "otis_mock_aime",
   "score": 0.6666666666666666,
   "source_value_text": "0.6666666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-22"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "gpqa_diamond",
   "score": 0.7121212121212122,
   "source_value_text": "0.7121212121212122",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-22"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7310605833333333,
   "source_value_text": "0.7310605833333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "simpleqa_verified",
   "score": 0.132,
   "source_value_text": "0.132",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "bfcl_overall",
   "score": 0.687,
   "source_value_text": "68.70%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "claude-haiku-4-5",
   "benchmark": "or_tau2_airline",
   "score": 0.6657576363636364,
   "source_value_text": "0.6657576363636364",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-3-7-sonnet",
   "benchmark": "math_level_5",
   "score": 0.9116314199395771,
   "source_value_text": "0.9116314199395771",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-13"
  },
  {
   "model": "claude-3-7-sonnet",
   "benchmark": "otis_mock_aime",
   "score": 0.5777777777777777,
   "source_value_text": "0.5777777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-13"
  },
  {
   "model": "claude-3-7-sonnet",
   "benchmark": "aider_polyglot",
   "score": 0.649,
   "source_value_text": "64.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-02-24"
  },
  {
   "model": "claude-3-7-sonnet",
   "benchmark": "swe_bench_verified",
   "score": 0.609504132231405,
   "source_value_text": "0.609504132231405",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-04"
  },
  {
   "model": "claude-3-7-sonnet",
   "benchmark": "gpqa_diamond",
   "score": 0.7973484848484849,
   "source_value_text": "0.7973484848484849",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-26"
  },
  {
   "model": "claude-opus-4",
   "benchmark": "math_level_5",
   "score": 0.850453172205438,
   "source_value_text": "0.850453172205438",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-22"
  },
  {
   "model": "claude-opus-4",
   "benchmark": "otis_mock_aime",
   "score": 0.6444444444444445,
   "source_value_text": "0.6444444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-28"
  },
  {
   "model": "claude-opus-4",
   "benchmark": "aider_polyglot",
   "score": 0.72,
   "source_value_text": "72.0",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-05-25"
  },
  {
   "model": "claude-opus-4",
   "benchmark": "swe_bench_verified",
   "score": 0.7066115702479339,
   "source_value_text": "0.7066115702479339",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-06"
  },
  {
   "model": "claude-opus-4",
   "benchmark": "gpqa_diamond",
   "score": 0.7626262626262627,
   "source_value_text": "0.7626262626262627",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-22"
  },
  {
   "model": "claude-sonnet-4",
   "benchmark": "math_level_5",
   "score": 0.8436555891238671,
   "source_value_text": "0.8436555891238671",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-22"
  },
  {
   "model": "claude-sonnet-4",
   "benchmark": "otis_mock_aime",
   "score": 0.7111111111111111,
   "source_value_text": "0.7111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-22"
  },
  {
   "model": "claude-sonnet-4",
   "benchmark": "aider_polyglot",
   "score": 0.613,
   "source_value_text": "61.3",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-05-24"
  },
  {
   "model": "claude-sonnet-4",
   "benchmark": "gpqa_diamond",
   "score": 0.7918781725888325,
   "source_value_text": "0.7918781725888325",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-26"
  },
  {
   "model": "claude-sonnet-4",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7672559999999999,
   "source_value_text": "0.7672559999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "claude-sonnet-4",
   "benchmark": "or_tau2_airline",
   "score": 0.64,
   "source_value_text": "0.64",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "qwen3-32b",
   "benchmark": "otis_mock_aime",
   "score": 0.6694444444444444,
   "source_value_text": "0.6694444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-32b",
   "benchmark": "aider_polyglot",
   "score": 0.4,
   "source_value_text": "40.0",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-05-08"
  },
  {
   "model": "qwen3-32b",
   "benchmark": "gpqa_diamond",
   "score": 0.6571969696969697,
   "source_value_text": "0.6571969696969697",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-32b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.5572391666666667,
   "source_value_text": "0.5572391666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "qwen3-32b",
   "benchmark": "bfcl_overall",
   "score": 0.48710000000000003,
   "source_value_text": "48.71%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-32b",
   "benchmark": "or_tau2_airline",
   "score": 0.4516665,
   "source_value_text": "0.4516665",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-09"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "chess_puzzles",
   "score": 0.01,
   "source_value_text": "0.01",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "mystery_game_puzzles",
   "score": 0.07,
   "source_value_text": "0.07",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "frontiermath_t13",
   "score": 0.06666666666666667,
   "source_value_text": "0.06666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "math_level_5",
   "score": 0.8729229607250756,
   "source_value_text": "0.8729229607250756",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.44722222222222224,
   "source_value_text": "0.44722222222222224",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "aider_polyglot",
   "score": 0.324,
   "source_value_text": "32.4",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.6584595959595959,
   "source_value_text": "0.6584595959595959",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6473065000000001,
   "source_value_text": "0.6473065000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "simpleqa_verified",
   "score": 0.127,
   "source_value_text": "0.127",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "bfcl_overall",
   "score": 0.5045000000000001,
   "source_value_text": "50.45%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-4-1-mini",
   "benchmark": "or_tau2_airline",
   "score": 0.44333325,
   "source_value_text": "0.44333325",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "chess_puzzles",
   "score": 0.03,
   "source_value_text": "0.03",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "frontiermath_t13",
   "score": 0.05964912280701754,
   "source_value_text": "0.05964912280701754",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "math_level_5",
   "score": 0.8300604229607251,
   "source_value_text": "0.8300604229607251",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "otis_mock_aime",
   "score": 0.38333333333333336,
   "source_value_text": "0.38333333333333336",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "aider_polyglot",
   "score": 0.524,
   "source_value_text": "52.4",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "swe_bench_verified",
   "score": 0.48541666666666666,
   "source_value_text": "0.48541666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-08"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "gpqa_diamond",
   "score": 0.6691919191919192,
   "source_value_text": "0.6691919191919192",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6607744999999999,
   "source_value_text": "0.6607744999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "simpleqa_verified",
   "score": 0.311,
   "source_value_text": "0.311",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "gpt-4-1",
   "benchmark": "bfcl_overall",
   "score": 0.5396,
   "source_value_text": "53.96%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-4-1",
   "benchmark": "or_tau2_airline",
   "score": 0.5083332500000001,
   "source_value_text": "0.5083332500000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "math_level_5",
   "score": 0.7301737160120846,
   "source_value_text": "0.7301737160120846",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-08"
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "otis_mock_aime",
   "score": 0.20555555555555555,
   "source_value_text": "0.20555555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-08"
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "aider_polyglot",
   "score": 0.156,
   "source_value_text": "15.6",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-06"
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "gpqa_diamond",
   "score": 0.6698232323232324,
   "source_value_text": "0.6698232323232324",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-08"
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6478116,
   "source_value_text": "0.6478116",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "bfcl_overall",
   "score": 0.3729,
   "source_value_text": "37.29%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "llama-4-maverick",
   "benchmark": "or_tau2_airline",
   "score": 0.43089426016260163,
   "source_value_text": "0.43089426016260163",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "gemma-3-27b",
   "benchmark": "math_level_5",
   "score": 0.740370090634441,
   "source_value_text": "0.740370090634441",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-13"
  },
  {
   "model": "gemma-3-27b",
   "benchmark": "otis_mock_aime",
   "score": 0.225,
   "source_value_text": "0.225",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemma-3-27b",
   "benchmark": "aider_polyglot",
   "score": 0.049,
   "source_value_text": "4.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-03-15"
  },
  {
   "model": "gemma-3-27b",
   "benchmark": "gpqa_diamond",
   "score": 0.4772727272727273,
   "source_value_text": "0.4772727272727273",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemma-3-27b",
   "benchmark": "bfcl_overall",
   "score": 0.29469999999999996,
   "source_value_text": "29.47%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "chess_puzzles",
   "score": 0.0,
   "source_value_text": "0.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "math_level_5",
   "score": 0.6999622356495468,
   "source_value_text": "0.6999622356495468",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "otis_mock_aime",
   "score": 0.28888888888888886,
   "source_value_text": "0.28888888888888886",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "aider_polyglot",
   "score": 0.08900000000000001,
   "source_value_text": "8.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "gpqa_diamond",
   "score": 0.48926767676767674,
   "source_value_text": "0.48926767676767674",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-14"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "or_gpqa_diamond",
   "score": 0.4966332500000001,
   "source_value_text": "0.4966332500000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "simpleqa_verified",
   "score": 0.06,
   "source_value_text": "0.06",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "bfcl_overall",
   "score": 0.33049999999999996,
   "source_value_text": "33.05%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-4-1-nano",
   "benchmark": "or_tau2_airline",
   "score": 0.11833325,
   "source_value_text": "0.11833325",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "chess_puzzles",
   "score": 0.06,
   "source_value_text": "0.06",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "frontiermath_t13",
   "score": 0.0035087719298245615,
   "source_value_text": "0.0035087719298245615",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "math_level_5",
   "score": 0.5327605740181269,
   "source_value_text": "0.5327605740181269",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.06388888888888888,
   "source_value_text": "0.06388888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "aider_polyglot",
   "score": 0.231,
   "source_value_text": "23.1",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-30"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.49210858585858586,
   "source_value_text": "0.49210858585858586",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "or_gpqa_diamond",
   "score": 0.5280586666666666,
   "source_value_text": "0.5280586666666666",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "simpleqa_verified",
   "score": 0.26,
   "source_value_text": "0.26",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "gpt-4o-aug-2024",
   "benchmark": "or_tau2_airline",
   "score": 0.43,
   "source_value_text": "0.43",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4o-nov-2024",
   "benchmark": "math_level_5",
   "score": 0.49773413897280966,
   "source_value_text": "0.49773413897280966",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-05"
  },
  {
   "model": "gpt-4o-nov-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.0625,
   "source_value_text": "0.0625",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gpt-4o-nov-2024",
   "benchmark": "aider_polyglot",
   "score": 0.182,
   "source_value_text": "18.2",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-30"
  },
  {
   "model": "gpt-4o-nov-2024",
   "benchmark": "swe_bench_verified",
   "score": 0.30991735537190085,
   "source_value_text": "0.30991735537190085",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-11"
  },
  {
   "model": "gpt-4o-nov-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.4788510101010101,
   "source_value_text": "0.4788510101010101",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-05"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "chess_puzzles",
   "score": 0.03,
   "source_value_text": "0.03",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "mystery_game_puzzles",
   "score": 0.12,
   "source_value_text": "0.12",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "frontiermath_t13",
   "score": 0.007017543859649123,
   "source_value_text": "0.007017543859649123",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "math_level_5",
   "score": 0.5263406344410876,
   "source_value_text": "0.5263406344410876",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.06944444444444445,
   "source_value_text": "0.06944444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-07-30"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "aider_polyglot",
   "score": 0.036000000000000004,
   "source_value_text": "3.6",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-21"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.37720959595959597,
   "source_value_text": "0.37720959595959597",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "or_gpqa_diamond",
   "score": 0.4305555,
   "source_value_text": "0.4305555",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "simpleqa_verified",
   "score": 0.083,
   "source_value_text": "0.083",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "gpt-4o-mini",
   "benchmark": "or_tau2_airline",
   "score": 0.28000025,
   "source_value_text": "0.28000025",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "llama-3-3-70b",
   "benchmark": "math_level_5",
   "score": 0.41597432024169184,
   "source_value_text": "0.41597432024169184",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-3-70b",
   "benchmark": "otis_mock_aime",
   "score": 0.05138888888888889,
   "source_value_text": "0.05138888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "llama-3-3-70b",
   "benchmark": "gpqa_diamond",
   "score": 0.4744318181818182,
   "source_value_text": "0.4744318181818182",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-3-70b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.4753785,
   "source_value_text": "0.4753785",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "llama-3-3-70b",
   "benchmark": "bfcl_overall",
   "score": 0.319,
   "source_value_text": "31.90%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "llama-3-3-70b",
   "benchmark": "or_tau2_airline",
   "score": 0.313333,
   "source_value_text": "0.313333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-10-04"
  },
  {
   "model": "gpt-5-5-pro",
   "benchmark": "chess_puzzles",
   "score": 0.62,
   "source_value_text": "0.62",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-5-pro",
   "benchmark": "frontiermath_t4",
   "score": 0.7804878048780488,
   "source_value_text": "0.7804878048780488",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-5-pro",
   "benchmark": "frontiermath_t13",
   "score": 0.8771929824561403,
   "source_value_text": "0.8771929824561403",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-5-pro",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-24"
  },
  {
   "model": "gpt-5-5-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.9392135642135643,
   "source_value_text": "0.9392135642135643",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-24"
  },
  {
   "model": "ai-co-mathematician",
   "benchmark": "frontiermath_t4",
   "score": 0.756,
   "source_value_text": "0.756",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "muse-wonder",
   "benchmark": "furniture_assembly",
   "score": 0.5416666666666666,
   "source_value_text": "0.5416666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "muse-wonder",
   "benchmark": "mystery_game_puzzles",
   "score": 0.51,
   "source_value_text": "0.51",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "muse-wonder",
   "benchmark": "frontiermath_t4",
   "score": 0.7073170731707317,
   "source_value_text": "0.7073170731707317",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "muse-wonder",
   "benchmark": "frontiermath_t13",
   "score": 0.8210526315789474,
   "source_value_text": "0.8210526315789474",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "muse-wonder",
   "benchmark": "otis_mock_aime",
   "score": 0.9944444444444445,
   "source_value_text": "0.9944444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "gpt-5-4-pro",
   "benchmark": "chess_puzzles",
   "score": 0.59,
   "source_value_text": "0.59",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-4-pro",
   "benchmark": "frontiermath_t4",
   "score": 0.5853658536585366,
   "source_value_text": "0.5853658536585366",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "gpt-5-4-pro",
   "benchmark": "frontiermath_t13",
   "score": 0.8245614035087719,
   "source_value_text": "0.8245614035087719",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "gpt-5-4-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.946,
   "source_value_text": "0.946",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-03-20"
  },
  {
   "model": "gpt-5-4-pro",
   "benchmark": "simpleqa_verified",
   "score": 0.463,
   "source_value_text": "0.463",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-2-pro",
   "benchmark": "chess_puzzles",
   "score": 0.6,
   "source_value_text": "0.6",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-2-pro",
   "benchmark": "frontiermath_t4",
   "score": 0.46,
   "source_value_text": "0.46",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "gpt-5-2-pro",
   "benchmark": "frontiermath_t13",
   "score": 0.74,
   "source_value_text": "0.74",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-13"
  },
  {
   "model": "mimo-v2-6-pro",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8989897499999999,
   "source_value_text": "0.8989897499999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "mimo-v2-6-pro",
   "benchmark": "or_tau2_airline",
   "score": 0.7983335000000001,
   "source_value_text": "0.7983335000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "chess_puzzles",
   "score": 0.46,
   "source_value_text": "0.46",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "mystery_game_puzzles",
   "score": 0.38,
   "source_value_text": "0.38",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-05"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "frontiermath_t4",
   "score": 0.4634146341463415,
   "source_value_text": "0.4634146341463415",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-04"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "frontiermath_t13",
   "score": 0.7473684210526316,
   "source_value_text": "0.7473684210526316",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-04"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "otis_mock_aime",
   "score": 0.9944444444444445,
   "source_value_text": "0.9944444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-04"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "gpqa_diamond",
   "score": 0.9267676767676768,
   "source_value_text": "0.9267676767676768",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-04"
  },
  {
   "model": "qwen-3-8-max",
   "benchmark": "simpleqa_verified",
   "score": 0.458,
   "source_value_text": "0.458",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "deepseek-v4-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8989900000000001,
   "source_value_text": "0.8989900000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-08"
  },
  {
   "model": "deepseek-v4-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.7755553333333334,
   "source_value_text": "0.7755553333333334",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-08"
  },
  {
   "model": "tencent-hy4-preview",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8973063333333335,
   "source_value_text": "0.8973063333333335",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "tencent-hy4-preview",
   "benchmark": "or_tau2_airline",
   "score": 0.7777776666666667,
   "source_value_text": "0.7777776666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "muse-spark-1-3",
   "benchmark": "chess_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "muse-spark-1-3",
   "benchmark": "mystery_game_puzzles",
   "score": 0.25,
   "source_value_text": "0.25",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-18"
  },
  {
   "model": "muse-spark-1-3",
   "benchmark": "frontiermath_t4",
   "score": 0.4634146341463415,
   "source_value_text": "0.4634146341463415",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-18"
  },
  {
   "model": "muse-spark-1-3",
   "benchmark": "frontiermath_t13",
   "score": 0.743859649122807,
   "source_value_text": "0.743859649122807",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-16"
  },
  {
   "model": "muse-spark-1-3",
   "benchmark": "otis_mock_aime",
   "score": 0.9916666666666667,
   "source_value_text": "0.9916666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-17"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "chess_puzzles",
   "score": 0.37,
   "source_value_text": "0.37",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "furniture_assembly",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-10"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "frontiermath_t4",
   "score": 0.34146341463414637,
   "source_value_text": "0.34146341463414637",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "frontiermath_t13",
   "score": 0.656140350877193,
   "source_value_text": "0.656140350877193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "otis_mock_aime",
   "score": 1.0,
   "source_value_text": "1.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "gpqa_diamond",
   "score": 0.922979797979798,
   "source_value_text": "0.922979797979798",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "qwen3-8-max-0902",
   "benchmark": "simpleqa_verified",
   "score": 0.473,
   "source_value_text": "0.473",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-02"
  },
  {
   "model": "muse-spark-1-1",
   "benchmark": "chess_puzzles",
   "score": 0.34,
   "source_value_text": "0.34",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "muse-spark-1-1",
   "benchmark": "simpleqa_verified",
   "score": 0.5778894472361809,
   "source_value_text": "0.5778894472361809",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "qwen3-8-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8962022556962025,
   "source_value_text": "0.8962022556962025",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-10"
  },
  {
   "model": "qwen3-8-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.7,
   "source_value_text": "0.7",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "grok-4-5",
   "benchmark": "chess_puzzles",
   "score": 0.35,
   "source_value_text": "0.35",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "grok-4-5",
   "benchmark": "furniture_assembly",
   "score": 0.225,
   "source_value_text": "0.225",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-24"
  },
  {
   "model": "grok-4-5",
   "benchmark": "frontiermath_t4",
   "score": 0.24390243902439024,
   "source_value_text": "0.24390243902439024",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "grok-4-5",
   "benchmark": "frontiermath_t13",
   "score": 0.5719298245614035,
   "source_value_text": "0.5719298245614035",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-09"
  },
  {
   "model": "grok-4-5",
   "benchmark": "otis_mock_aime",
   "score": 0.9777777777777777,
   "source_value_text": "0.9777777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-08"
  },
  {
   "model": "grok-4-5",
   "benchmark": "gpqa_diamond",
   "score": 0.9343434343434344,
   "source_value_text": "0.9343434343434344",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-08"
  },
  {
   "model": "grok-4-5",
   "benchmark": "simpleqa_verified",
   "score": 0.483,
   "source_value_text": "0.483",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-5-pro",
   "benchmark": "chess_puzzles",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "gpt-5-pro",
   "benchmark": "frontiermath_t4",
   "score": 0.1951219512195122,
   "source_value_text": "0.1951219512195122",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "gpt-5-pro",
   "benchmark": "frontiermath_t13",
   "score": 0.5578947368421052,
   "source_value_text": "0.5578947368421052",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-12"
  },
  {
   "model": "grok-4-7",
   "benchmark": "chess_puzzles",
   "score": 0.41,
   "source_value_text": "0.41",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "grok-4-7",
   "benchmark": "furniture_assembly",
   "score": 0.20833333333333334,
   "source_value_text": "0.20833333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-24"
  },
  {
   "model": "grok-4-7",
   "benchmark": "mystery_game_puzzles",
   "score": 0.29,
   "source_value_text": "0.29",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "grok-4-7",
   "benchmark": "frontiermath_t4",
   "score": 0.17073170731707318,
   "source_value_text": "0.17073170731707318",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "grok-4-7",
   "benchmark": "frontiermath_t13",
   "score": 0.5298245614035088,
   "source_value_text": "0.5298245614035088",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "grok-4-7",
   "benchmark": "otis_mock_aime",
   "score": 0.9805555555555555,
   "source_value_text": "0.9805555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "grok-4-7",
   "benchmark": "gpqa_diamond",
   "score": 0.92739898989899,
   "source_value_text": "0.92739898989899",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "grok-4-7",
   "benchmark": "simpleqa_verified",
   "score": 0.56,
   "source_value_text": "0.56",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-09-22"
  },
  {
   "model": "gemini-3-pro",
   "benchmark": "otis_mock_aime",
   "score": 0.9138888888888889,
   "source_value_text": "0.9138888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-19"
  },
  {
   "model": "gemini-3-pro",
   "benchmark": "swe_bench_verified",
   "score": 0.7293388429752066,
   "source_value_text": "0.7293388429752066",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-13"
  },
  {
   "model": "gemini-3-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.9261363636363636,
   "source_value_text": "0.9261363636363636",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-11-19"
  },
  {
   "model": "gemini-3-pro",
   "benchmark": "bfcl_overall",
   "score": 0.7251000000000001,
   "source_value_text": "72.51%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-5-3-codex",
   "benchmark": "swe_bench_verified",
   "score": 0.7479338842975206,
   "source_value_text": "0.7479338842975206",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-25"
  },
  {
   "model": "gpt-5-3-codex",
   "benchmark": "or_tau2_airline",
   "score": 0.753333,
   "source_value_text": "0.753333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "chess_puzzles",
   "score": 0.33,
   "source_value_text": "0.33",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "furniture_assembly",
   "score": 0.20833333333333334,
   "source_value_text": "0.20833333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#furniture_assembly.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/furniture-assembly",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-14"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "mystery_game_puzzles",
   "score": 0.17,
   "source_value_text": "0.17",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "frontiermath_t4",
   "score": 0.2682926829268293,
   "source_value_text": "0.2682926829268293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "frontiermath_t13",
   "score": 0.5859649122807018,
   "source_value_text": "0.5859649122807018",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "otis_mock_aime",
   "score": 0.9833333333333333,
   "source_value_text": "0.9833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "gpqa_diamond",
   "score": 0.8806818181818182,
   "source_value_text": "0.8806818181818182",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-12"
  },
  {
   "model": "mistral-large-4",
   "benchmark": "simpleqa_verified",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-09"
  },
  {
   "model": "mimo-v2-5-pro",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8550657272727274,
   "source_value_text": "0.8550657272727274",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "mimo-v2-5-pro",
   "benchmark": "or_tau2_airline",
   "score": 0.743889,
   "source_value_text": "0.743889",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "muse-spark-1-2",
   "benchmark": "chess_puzzles",
   "score": 0.27,
   "source_value_text": "0.27",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "muse-spark-1-2",
   "benchmark": "simpleqa_verified",
   "score": 0.6030150753768844,
   "source_value_text": "0.6030150753768844",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "minimax-m2-7",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8570973081339713,
   "source_value_text": "0.8570973081339713",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "minimax-m2-7",
   "benchmark": "or_tau2_airline",
   "score": 0.7203921764705883,
   "source_value_text": "0.7203921764705883",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "minimax-m2-1",
   "benchmark": "or_gpqa_diamond",
   "score": 0.868687,
   "source_value_text": "0.868687",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "minimax-m2-1",
   "benchmark": "or_tau2_airline",
   "score": 0.6783334999999999,
   "source_value_text": "0.6783334999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "qwen-3-8-27b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.83367,
   "source_value_text": "0.83367",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen-3-8-27b",
   "benchmark": "or_tau2_airline",
   "score": 0.7592592222222222,
   "source_value_text": "0.7592592222222222",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "o3-pro",
   "benchmark": "aider_polyglot",
   "score": 0.8490000000000001,
   "source_value_text": "84.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-06-28"
  },
  {
   "model": "minimax-m2-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8539973475609757,
   "source_value_text": "0.8539973475609757",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "minimax-m2-5",
   "benchmark": "or_tau2_airline",
   "score": 0.6905556666666667,
   "source_value_text": "0.6905556666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "muse-spark",
   "benchmark": "otis_mock_aime",
   "score": 0.889,
   "source_value_text": "0.889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-08"
  },
  {
   "model": "muse-spark",
   "benchmark": "gpqa_diamond",
   "score": 0.898,
   "source_value_text": "0.898",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-04-08"
  },
  {
   "model": "qwen-3-6-max-preview",
   "benchmark": "mystery_game_puzzles",
   "score": 0.19,
   "source_value_text": "0.19",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen-3-6-max-preview",
   "benchmark": "otis_mock_aime",
   "score": 0.9111111111111111,
   "source_value_text": "0.9111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-max-preview",
   "benchmark": "swe_bench_verified",
   "score": 0.7665289256198347,
   "source_value_text": "0.7665289256198347",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-28"
  },
  {
   "model": "qwen-3-6-max-preview",
   "benchmark": "gpqa_diamond",
   "score": 0.8737373737373737,
   "source_value_text": "0.8737373737373737",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-max-preview",
   "benchmark": "simpleqa_verified",
   "score": 0.5200400801603207,
   "source_value_text": "0.5200400801603207",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "mimo-v2-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8239975454545455,
   "source_value_text": "0.8239975454545455",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "mimo-v2-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7372220833333334,
   "source_value_text": "0.7372220833333334",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "grok-4-3-beta",
   "benchmark": "chess_puzzles",
   "score": 0.18,
   "source_value_text": "0.18",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "grok-4-3-beta",
   "benchmark": "frontiermath_t4",
   "score": 0.14634146341463414,
   "source_value_text": "0.14634146341463414",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "grok-4-3-beta",
   "benchmark": "frontiermath_t13",
   "score": 0.4280701754385965,
   "source_value_text": "0.4280701754385965",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "grok-4-3-beta",
   "benchmark": "otis_mock_aime",
   "score": 0.9333333333333333,
   "source_value_text": "0.9333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "grok-4-3-beta",
   "benchmark": "gpqa_diamond",
   "score": 0.8882575757575758,
   "source_value_text": "0.8882575757575758",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-17"
  },
  {
   "model": "grok-4-3-beta",
   "benchmark": "simpleqa_verified",
   "score": 0.332,
   "source_value_text": "0.332",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "glm-4-6",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7646065075987841,
   "source_value_text": "0.7646065075987841",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "glm-4-6",
   "benchmark": "bfcl_overall",
   "score": 0.7238,
   "source_value_text": "72.38%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "glm-4-6",
   "benchmark": "or_tau2_airline",
   "score": 0.7344445,
   "source_value_text": "0.7344445",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-4-5-air",
   "benchmark": "or_tau2_airline",
   "score": 0.68,
   "source_value_text": "0.68",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "deepseek-v3-2",
   "benchmark": "or_gpqa_diamond",
   "score": 0.802714625,
   "source_value_text": "0.802714625",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-v3-2",
   "benchmark": "or_tau2_airline",
   "score": 0.7193332,
   "source_value_text": "0.7193332",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-04"
  },
  {
   "model": "nemotron-3-ultra",
   "benchmark": "mystery_game_puzzles",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "nemotron-3-ultra",
   "benchmark": "otis_mock_aime",
   "score": 0.8666666666666667,
   "source_value_text": "0.8666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "nemotron-3-ultra",
   "benchmark": "gpqa_diamond",
   "score": 0.8535353535353535,
   "source_value_text": "0.8535353535353535",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-10"
  },
  {
   "model": "nemotron-3-ultra",
   "benchmark": "or_tau2_airline",
   "score": 0.72,
   "source_value_text": "0.72",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-04"
  },
  {
   "model": "grok-4",
   "benchmark": "otis_mock_aime",
   "score": 0.84,
   "source_value_text": "0.84",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": ""
  },
  {
   "model": "grok-4",
   "benchmark": "aider_polyglot",
   "score": 0.7959999999999999,
   "source_value_text": "79.6",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-07-11"
  },
  {
   "model": "grok-4",
   "benchmark": "gpqa_diamond",
   "score": 0.87,
   "source_value_text": "0.87",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": ""
  },
  {
   "model": "grok-4",
   "benchmark": "bfcl_overall",
   "score": 0.6297,
   "source_value_text": "62.97%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "chess_puzzles",
   "score": 0.14,
   "source_value_text": "0.14",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "mystery_game_puzzles",
   "score": 0.12,
   "source_value_text": "0.12",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "frontiermath_t13",
   "score": 0.3824561403508772,
   "source_value_text": "0.3824561403508772",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "otis_mock_aime",
   "score": 0.9333333333333333,
   "source_value_text": "0.9333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "swe_bench_verified",
   "score": 0.5785123966942148,
   "source_value_text": "0.5785123966942148",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-05-14"
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "gpqa_diamond",
   "score": 0.8838383838383839,
   "source_value_text": "0.8838383838383839",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-plus",
   "benchmark": "simpleqa_verified",
   "score": 0.44144144144144143,
   "source_value_text": "0.44144144144144143",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "kimi-k2-thinking",
   "benchmark": "or_gpqa_diamond",
   "score": 0.796581,
   "source_value_text": "0.796581",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "kimi-k2-thinking",
   "benchmark": "or_tau2_airline",
   "score": 0.695,
   "source_value_text": "0.695",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "qwen3-5-122b-a10b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.17,
   "source_value_text": "0.17",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-5-122b-a10b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8073469346289753,
   "source_value_text": "0.8073469346289753",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-5-122b-a10b",
   "benchmark": "or_tau2_airline",
   "score": 0.7471783002257337,
   "source_value_text": "0.7471783002257337",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "mimo-v2-6-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.758698,
   "source_value_text": "0.758698",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "mimo-v2-6-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.76,
   "source_value_text": "0.76",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-v3-2-exp",
   "benchmark": "aider_polyglot",
   "score": 0.742,
   "source_value_text": "74.2",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-10-03"
  },
  {
   "model": "deepseek-v3-2-exp",
   "benchmark": "or_gpqa_diamond",
   "score": 0.8265989999999999,
   "source_value_text": "0.8265989999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "deepseek-v3-2-exp",
   "benchmark": "bfcl_overall",
   "score": 0.5672999999999999,
   "source_value_text": "56.73%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "deepseek-v3-2-exp",
   "benchmark": "or_tau2_airline",
   "score": 0.693333,
   "source_value_text": "0.693333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "glm-5",
   "benchmark": "otis_mock_aime",
   "score": 0.8,
   "source_value_text": "0.8",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-12"
  },
  {
   "model": "glm-5",
   "benchmark": "swe_bench_verified",
   "score": 0.7207792207792207,
   "source_value_text": "0.7207792207792207",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-15"
  },
  {
   "model": "glm-5",
   "benchmark": "gpqa_diamond",
   "score": 0.8781725888324873,
   "source_value_text": "0.8781725888324873",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-12"
  },
  {
   "model": "glm-5",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7285024173913045,
   "source_value_text": "0.7285024173913045",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-5",
   "benchmark": "or_tau2_airline",
   "score": 0.7422222222222222,
   "source_value_text": "0.7422222222222222",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemini-3-1-flash-lite",
   "benchmark": "frontiermath_t13",
   "score": 0.2771929824561403,
   "source_value_text": "0.2771929824561403",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "gemini-3-1-flash-lite",
   "benchmark": "otis_mock_aime",
   "score": 0.8,
   "source_value_text": "0.8",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemini-3-1-flash-lite",
   "benchmark": "gpqa_diamond",
   "score": 0.8181818181818182,
   "source_value_text": "0.8181818181818182",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "qwen-3-5-plus-hosted-397b-a17b",
   "benchmark": "chess_puzzles",
   "score": 0.08,
   "source_value_text": "0.08",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen-3-5-plus-hosted-397b-a17b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.17,
   "source_value_text": "0.17",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen-3-5-plus-hosted-397b-a17b",
   "benchmark": "otis_mock_aime",
   "score": 0.8666666666666667,
   "source_value_text": "0.8666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-5-plus-hosted-397b-a17b",
   "benchmark": "gpqa_diamond",
   "score": 0.8484848484848485,
   "source_value_text": "0.8484848484848485",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-5-plus-hosted-397b-a17b",
   "benchmark": "simpleqa_verified",
   "score": 0.2535070140280561,
   "source_value_text": "0.2535070140280561",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemini-2-5-pro-mar-2025",
   "benchmark": "math_level_5",
   "score": 0.9556268882175226,
   "source_value_text": "0.9556268882175226",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-07"
  },
  {
   "model": "gemini-2-5-pro-mar-2025",
   "benchmark": "aider_polyglot",
   "score": 0.7290000000000001,
   "source_value_text": "72.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-12"
  },
  {
   "model": "gemini-2-5-pro-mar-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.8383838383838383,
   "source_value_text": "0.8383838383838383",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-31"
  },
  {
   "model": "qwen-3-6-flash",
   "benchmark": "chess_puzzles",
   "score": 0.12,
   "source_value_text": "0.12",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen-3-6-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.18,
   "source_value_text": "0.18",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen-3-6-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.22456140350877193,
   "source_value_text": "0.22456140350877193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "qwen-3-6-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.8444444444444444,
   "source_value_text": "0.8444444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.8333333333333334,
   "source_value_text": "0.8333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-6-flash",
   "benchmark": "simpleqa_verified",
   "score": 0.15915915915915915,
   "source_value_text": "0.15915915915915915",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-7-flash",
   "benchmark": "mystery_game_puzzles",
   "score": 0.15,
   "source_value_text": "0.15",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-7-flash",
   "benchmark": "frontiermath_t13",
   "score": 0.19298245614035087,
   "source_value_text": "0.19298245614035087",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "qwen3-7-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.8666666666666667,
   "source_value_text": "0.8666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen3-7-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.8232323232323232,
   "source_value_text": "0.8232323232323232",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "gpt-5-5-instant",
   "benchmark": "frontiermath_t4",
   "score": 0.024390243902439025,
   "source_value_text": "0.024390243902439025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gpt-5-5-instant",
   "benchmark": "frontiermath_t13",
   "score": 0.2631578947368421,
   "source_value_text": "0.2631578947368421",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gpt-5-5-instant",
   "benchmark": "otis_mock_aime",
   "score": 0.6805555555555556,
   "source_value_text": "0.6805555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "gpt-5-5-instant",
   "benchmark": "gpqa_diamond",
   "score": 0.8251262626262627,
   "source_value_text": "0.8251262626262627",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-02"
  },
  {
   "model": "qwen-3-5-flash-hosted-35b-a3b",
   "benchmark": "chess_puzzles",
   "score": 0.14,
   "source_value_text": "0.14",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "qwen-3-5-flash-hosted-35b-a3b",
   "benchmark": "mystery_game_puzzles",
   "score": 0.2,
   "source_value_text": "0.2",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen-3-5-flash-hosted-35b-a3b",
   "benchmark": "frontiermath_t13",
   "score": 0.1824561403508772,
   "source_value_text": "0.1824561403508772",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen-3-5-flash-hosted-35b-a3b",
   "benchmark": "otis_mock_aime",
   "score": 0.8444444444444444,
   "source_value_text": "0.8444444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-5-flash-hosted-35b-a3b",
   "benchmark": "gpqa_diamond",
   "score": 0.8232323232323232,
   "source_value_text": "0.8232323232323232",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "qwen-3-5-flash-hosted-35b-a3b",
   "benchmark": "simpleqa_verified",
   "score": 0.2032032032032032,
   "source_value_text": "0.2032032032032032",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-max",
   "benchmark": "mystery_game_puzzles",
   "score": 0.05,
   "source_value_text": "0.05",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-max",
   "benchmark": "frontiermath_t13",
   "score": 0.18947368421052632,
   "source_value_text": "0.18947368421052632",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-max",
   "benchmark": "math_level_5",
   "score": 0.9712990936555891,
   "source_value_text": "0.9712990936555891",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-09"
  },
  {
   "model": "qwen3-max",
   "benchmark": "otis_mock_aime",
   "score": 0.7333333333333333,
   "source_value_text": "0.7333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-06"
  },
  {
   "model": "qwen3-max",
   "benchmark": "gpqa_diamond",
   "score": 0.726010101010101,
   "source_value_text": "0.726010101010101",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-06"
  },
  {
   "model": "qwen3-max",
   "benchmark": "simpleqa_verified",
   "score": 0.4874874874874875,
   "source_value_text": "0.4874874874874875",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemma-4-26b-a4b",
   "benchmark": "otis_mock_aime",
   "score": 0.8222222222222222,
   "source_value_text": "0.8222222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-4-26b-a4b",
   "benchmark": "gpqa_diamond",
   "score": 0.7323232323232324,
   "source_value_text": "0.7323232323232324",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemma-4-26b-a4b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7762077565217391,
   "source_value_text": "0.7762077565217391",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemma-4-26b-a4b",
   "benchmark": "or_tau2_airline",
   "score": 0.5733334999999999,
   "source_value_text": "0.5733334999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-5-35b-a3b",
   "benchmark": "otis_mock_aime",
   "score": 0.7,
   "source_value_text": "0.7",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-5-35b-a3b",
   "benchmark": "gpqa_diamond",
   "score": 0.8345959595959596,
   "source_value_text": "0.8345959595959596",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-5-35b-a3b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7500000000000001,
   "source_value_text": "0.7500000000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-5-35b-a3b",
   "benchmark": "or_tau2_airline",
   "score": 0.7351189375,
   "source_value_text": "0.7351189375",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "claude-opus-4-1",
   "benchmark": "mystery_game_puzzles",
   "score": 0.21,
   "source_value_text": "0.21",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-07-25"
  },
  {
   "model": "claude-opus-4-1",
   "benchmark": "frontiermath_t4",
   "score": 0.024390243902439025,
   "source_value_text": "0.024390243902439025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tier_4_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tier-4-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-opus-4-1",
   "benchmark": "frontiermath_t13",
   "score": 0.12631578947368421,
   "source_value_text": "0.12631578947368421",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-06-11"
  },
  {
   "model": "claude-opus-4-1",
   "benchmark": "otis_mock_aime",
   "score": 0.6888888888888889,
   "source_value_text": "0.6888888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-08-05"
  },
  {
   "model": "claude-opus-4-1",
   "benchmark": "swe_bench_verified",
   "score": 0.7334710743801653,
   "source_value_text": "0.7334710743801653",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#swe_bench_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/swe-bench-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-02-11"
  },
  {
   "model": "claude-opus-4-1",
   "benchmark": "gpqa_diamond",
   "score": 0.7727272727272727,
   "source_value_text": "0.7727272727272727",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-08-05"
  },
  {
   "model": "kimi-k2-sep-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7474749999999999,
   "source_value_text": "0.7474749999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "kimi-k2-sep-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.6366665,
   "source_value_text": "0.6366665",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "deepseek-r1-may-2025",
   "benchmark": "math_level_5",
   "score": 0.9663897280966768,
   "source_value_text": "0.9663897280966768",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-29"
  },
  {
   "model": "deepseek-r1-may-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.6638888888888889,
   "source_value_text": "0.6638888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-29"
  },
  {
   "model": "deepseek-r1-may-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.7632575757575758,
   "source_value_text": "0.7632575757575758",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-29"
  },
  {
   "model": "deepseek-r1-may-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7851099187817259,
   "source_value_text": "0.7851099187817259",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-r1-may-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.47874697986577186,
   "source_value_text": "0.47874697986577186",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemini-2-5-pro-may-2025",
   "benchmark": "math_level_5",
   "score": 0.9590256797583081,
   "source_value_text": "0.9590256797583081",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-08"
  },
  {
   "model": "gemini-2-5-pro-may-2025",
   "benchmark": "aider_polyglot",
   "score": 0.769,
   "source_value_text": "76.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-05-07"
  },
  {
   "model": "gemini-2-5-pro-may-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.6666666666666666,
   "source_value_text": "0.6666666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-06-03"
  },
  {
   "model": "gemma-4-31b-it",
   "benchmark": "otis_mock_aime",
   "score": 0.7333333333333333,
   "source_value_text": "0.7333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemma-4-31b-it",
   "benchmark": "gpqa_diamond",
   "score": 0.7575757575757576,
   "source_value_text": "0.7575757575757576",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "gemma-4-31b-it",
   "benchmark": "or_gpqa_diamond",
   "score": 0.81649825,
   "source_value_text": "0.81649825",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gemma-4-31b-it",
   "benchmark": "simpleqa_verified",
   "score": 0.104,
   "source_value_text": "0.104",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gemma-4-31b-it",
   "benchmark": "or_tau2_airline",
   "score": 0.756666625,
   "source_value_text": "0.756666625",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-coder-next",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7575756666666668,
   "source_value_text": "0.7575756666666668",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "qwen3-coder-next",
   "benchmark": "or_tau2_airline",
   "score": 0.5762463812316716,
   "source_value_text": "0.5762463812316716",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "o1",
   "benchmark": "chess_puzzles",
   "score": 0.1,
   "source_value_text": "0.1",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "o1",
   "benchmark": "frontiermath_t13",
   "score": 0.14736842105263157,
   "source_value_text": "0.14736842105263157",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "o1",
   "benchmark": "math_level_5",
   "score": 0.947129909365559,
   "source_value_text": "0.947129909365559",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-13"
  },
  {
   "model": "o1",
   "benchmark": "otis_mock_aime",
   "score": 0.7333333333333333,
   "source_value_text": "0.7333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-27"
  },
  {
   "model": "o1",
   "benchmark": "gpqa_diamond",
   "score": 0.7676767676767676,
   "source_value_text": "0.7676767676767676",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-13"
  },
  {
   "model": "o1",
   "benchmark": "simpleqa_verified",
   "score": 0.411,
   "source_value_text": "0.411",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "deepseek-v3-1-terminus",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7942762,
   "source_value_text": "0.7942762",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "deepseek-v3-1-terminus",
   "benchmark": "or_tau2_airline",
   "score": 0.48266679999999995,
   "source_value_text": "0.48266679999999995",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "gemini-2-5-flash-jun-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.738309,
   "source_value_text": "0.738309",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-2-5-flash-jun-2025",
   "benchmark": "bfcl_overall",
   "score": 0.5624,
   "source_value_text": "56.24%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gemini-2-5-flash-jun-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.5511113333333334,
   "source_value_text": "0.5511113333333334",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "gemini-2-5-flash-may-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.7083333333333334,
   "source_value_text": "0.7083333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-20"
  },
  {
   "model": "gemini-2-5-flash-may-2025",
   "benchmark": "aider_polyglot",
   "score": 0.551,
   "source_value_text": "55.1",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-05-25"
  },
  {
   "model": "gemini-2-5-flash-apr-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.7305555555555555,
   "source_value_text": "0.7305555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-22"
  },
  {
   "model": "gemini-2-5-flash-apr-2025",
   "benchmark": "aider_polyglot",
   "score": 0.47100000000000003,
   "source_value_text": "47.1",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-20"
  },
  {
   "model": "qwen3-5-9b",
   "benchmark": "otis_mock_aime",
   "score": 0.6166666666666667,
   "source_value_text": "0.6166666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-5-9b",
   "benchmark": "gpqa_diamond",
   "score": 0.7897727272727273,
   "source_value_text": "0.7897727272727273",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-5-9b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7462993119730186,
   "source_value_text": "0.7462993119730186",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "qwen3-5-9b",
   "benchmark": "or_tau2_airline",
   "score": 0.650067849593496,
   "source_value_text": "0.650067849593496",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "grok-3-mini",
   "benchmark": "math_level_5",
   "score": 0.9093655589123867,
   "source_value_text": "0.9093655589123867",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "grok-3-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.7777777777777778,
   "source_value_text": "0.7777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "grok-3-mini",
   "benchmark": "aider_polyglot",
   "score": 0.493,
   "source_value_text": "49.3",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "grok-3-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.7626262626262627,
   "source_value_text": "0.7626262626262627",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "deepseek-v3-1",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7707982259154932,
   "source_value_text": "0.7707982259154932",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "deepseek-v3-1",
   "benchmark": "or_tau2_airline",
   "score": 0.47222181249999995,
   "source_value_text": "0.47222181249999995",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-15"
  },
  {
   "model": "qwen3-30b-a3b-thinking-jul-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.7027777777777777,
   "source_value_text": "0.7027777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-30b-a3b-thinking-jul-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.7007575757575758,
   "source_value_text": "0.7007575757575758",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "seed-oss-36b-instruct",
   "benchmark": "otis_mock_aime",
   "score": 0.675,
   "source_value_text": "0.675",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "seed-oss-36b-instruct",
   "benchmark": "gpqa_diamond",
   "score": 0.7152777777777778,
   "source_value_text": "0.7152777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "deepseek-r1",
   "benchmark": "math_level_5",
   "score": 0.9305135951661632,
   "source_value_text": "0.9305135951661632",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-31"
  },
  {
   "model": "deepseek-r1",
   "benchmark": "otis_mock_aime",
   "score": 0.5333333333333333,
   "source_value_text": "0.5333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-26"
  },
  {
   "model": "deepseek-r1",
   "benchmark": "aider_polyglot",
   "score": 0.569,
   "source_value_text": "56.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-01-20"
  },
  {
   "model": "deepseek-r1",
   "benchmark": "gpqa_diamond",
   "score": 0.7171717171717171,
   "source_value_text": "0.7171717171717171",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-26"
  },
  {
   "model": "qwen3-235b-a22b-instruct-jul-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7388167142857143,
   "source_value_text": "0.7388167142857143",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "qwen3-235b-a22b-instruct-jul-2025",
   "benchmark": "bfcl_overall",
   "score": 0.5215,
   "source_value_text": "52.15%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-235b-a22b-instruct-jul-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.3961905714285714,
   "source_value_text": "0.3961905714285714",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "grok-3",
   "benchmark": "math_level_5",
   "score": 0.8874622356495468,
   "source_value_text": "0.8874622356495468",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "grok-3",
   "benchmark": "otis_mock_aime",
   "score": 0.5555555555555556,
   "source_value_text": "0.5555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "grok-3",
   "benchmark": "aider_polyglot",
   "score": 0.5329999999999999,
   "source_value_text": "53.3",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-10"
  },
  {
   "model": "grok-3",
   "benchmark": "gpqa_diamond",
   "score": 0.7575757575757576,
   "source_value_text": "0.7575757575757576",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-26"
  },
  {
   "model": "qwen3-5-4b",
   "benchmark": "otis_mock_aime",
   "score": 0.5583333333333333,
   "source_value_text": "0.5583333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "mistral-small-4",
   "benchmark": "or_gpqa_diamond",
   "score": 0.7528056666666667,
   "source_value_text": "0.7528056666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-09-17"
  },
  {
   "model": "mistral-small-4",
   "benchmark": "or_tau2_airline",
   "score": 0.41142871428571426,
   "source_value_text": "0.41142871428571426",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-09-17"
  },
  {
   "model": "gpt-oss-20b",
   "benchmark": "otis_mock_aime",
   "score": 0.6527777777777778,
   "source_value_text": "0.6527777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-oss-20b",
   "benchmark": "gpqa_diamond",
   "score": 0.6079545454545454,
   "source_value_text": "0.6079545454545454",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-oss-20b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6311448,
   "source_value_text": "0.6311448",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:52:14Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "gpt-oss-20b",
   "benchmark": "or_tau2_airline",
   "score": 0.4672696511627907,
   "source_value_text": "0.4672696511627907",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "glm-4-7-flash",
   "benchmark": "otis_mock_aime",
   "score": 0.5833333333333334,
   "source_value_text": "0.5833333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "glm-4-7-flash",
   "benchmark": "gpqa_diamond",
   "score": 0.6054292929292929,
   "source_value_text": "0.6054292929292929",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "glm-4-7-flash",
   "benchmark": "or_gpqa_diamond",
   "score": 0.49197393750000007,
   "source_value_text": "0.49197393750000007",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "glm-4-7-flash",
   "benchmark": "or_tau2_airline",
   "score": 0.6833333333333333,
   "source_value_text": "0.6833333333333333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-30b-a3b",
   "benchmark": "otis_mock_aime",
   "score": 0.6277777777777778,
   "source_value_text": "0.6277777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-30b-a3b",
   "benchmark": "gpqa_diamond",
   "score": 0.6167929292929293,
   "source_value_text": "0.6167929292929293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-30b-a3b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.604377,
   "source_value_text": "0.604377",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "qwen3-30b-a3b",
   "benchmark": "or_tau2_airline",
   "score": 0.423333,
   "source_value_text": "0.423333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "cohere-command-a",
   "benchmark": "bfcl_overall",
   "score": 0.46490000000000004,
   "source_value_text": "46.49%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "deepseek-r1-distill-qwen-32b",
   "benchmark": "otis_mock_aime",
   "score": 0.5555555555555556,
   "source_value_text": "0.5555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "deepseek-r1-distill-qwen-32b",
   "benchmark": "gpqa_diamond",
   "score": 0.6414141414141414,
   "source_value_text": "0.6414141414141414",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-14b",
   "benchmark": "otis_mock_aime",
   "score": 0.6638888888888889,
   "source_value_text": "0.6638888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-14b",
   "benchmark": "gpqa_diamond",
   "score": 0.6376262626262627,
   "source_value_text": "0.6376262626262627",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-14b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.5331086666666667,
   "source_value_text": "0.5331086666666667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "qwen3-14b",
   "benchmark": "bfcl_overall",
   "score": 0.4103,
   "source_value_text": "41.03%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-14b",
   "benchmark": "or_tau2_airline",
   "score": 0.413333,
   "source_value_text": "0.413333",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "deepseek-r1-distill-llama-70b",
   "benchmark": "math_level_5",
   "score": 0.8989803625377644,
   "source_value_text": "0.8989803625377644",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-10"
  },
  {
   "model": "deepseek-r1-distill-llama-70b",
   "benchmark": "otis_mock_aime",
   "score": 0.5138888888888888,
   "source_value_text": "0.5138888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "deepseek-r1-distill-llama-70b",
   "benchmark": "gpqa_diamond",
   "score": 0.5574494949494949,
   "source_value_text": "0.5574494949494949",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-10"
  },
  {
   "model": "qwq-32b",
   "benchmark": "otis_mock_aime",
   "score": 0.5916666666666667,
   "source_value_text": "0.5916666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwq-32b",
   "benchmark": "aider_polyglot",
   "score": 0.209,
   "source_value_text": "20.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-03-06"
  },
  {
   "model": "qwq-32b",
   "benchmark": "gpqa_diamond",
   "score": 0.6534090909090909,
   "source_value_text": "0.6534090909090909",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "o1-mini",
   "benchmark": "math_level_5",
   "score": 0.8918051359516617,
   "source_value_text": "0.8918051359516617",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-13"
  },
  {
   "model": "o1-mini",
   "benchmark": "otis_mock_aime",
   "score": 0.46944444444444444,
   "source_value_text": "0.46944444444444444",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-06"
  },
  {
   "model": "o1-mini",
   "benchmark": "aider_polyglot",
   "score": 0.32899999999999996,
   "source_value_text": "32.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-22"
  },
  {
   "model": "o1-mini",
   "benchmark": "gpqa_diamond",
   "score": 0.6237373737373737,
   "source_value_text": "0.6237373737373737",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-13"
  },
  {
   "model": "qwen3-30b-a3b-instruct-jul-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.6222222222222222,
   "source_value_text": "0.6222222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-30b-a3b-instruct-jul-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.5561868686868687,
   "source_value_text": "0.5561868686868687",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-30b-a3b-instruct-jul-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6419754999999999,
   "source_value_text": "0.6419754999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "qwen3-30b-a3b-instruct-jul-2025",
   "benchmark": "bfcl_overall",
   "score": 0.4139,
   "source_value_text": "41.39%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-30b-a3b-instruct-jul-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.3695237142857143,
   "source_value_text": "0.3695237142857143",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "qwen3-8b",
   "benchmark": "otis_mock_aime",
   "score": 0.5611111111111111,
   "source_value_text": "0.5611111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-8b",
   "benchmark": "gpqa_diamond",
   "score": 0.5675505050505051,
   "source_value_text": "0.5675505050505051",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "qwen3-8b",
   "benchmark": "bfcl_overall",
   "score": 0.4257,
   "source_value_text": "42.57%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "deepseek-r1-0528-qwen3-8b",
   "benchmark": "otis_mock_aime",
   "score": 0.4388888888888889,
   "source_value_text": "0.4388888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "deepseek-r1-0528-qwen3-8b",
   "benchmark": "gpqa_diamond",
   "score": 0.0928030303030303,
   "source_value_text": "0.0928030303030303",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwq-plus",
   "benchmark": "gpqa_diamond",
   "score": 0.6540404040404041,
   "source_value_text": "0.6540404040404041",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-11"
  },
  {
   "model": "deepseek-r1-distill-qwen-14b",
   "benchmark": "math_level_5",
   "score": 0.8712235649546828,
   "source_value_text": "0.8712235649546828",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-11"
  },
  {
   "model": "deepseek-r1-distill-qwen-14b",
   "benchmark": "otis_mock_aime",
   "score": 0.5055555555555555,
   "source_value_text": "0.5055555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "deepseek-r1-distill-qwen-14b",
   "benchmark": "gpqa_diamond",
   "score": 0.44696969696969696,
   "source_value_text": "0.44696969696969696",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-10"
  },
  {
   "model": "gemini-2-0-pro",
   "benchmark": "math_level_5",
   "score": 0.8345921450151057,
   "source_value_text": "0.8345921450151057",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-06"
  },
  {
   "model": "gemini-2-0-pro",
   "benchmark": "aider_polyglot",
   "score": 0.35600000000000004,
   "source_value_text": "35.6",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-2-0-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.6565656565656566,
   "source_value_text": "0.6565656565656566",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-07"
  },
  {
   "model": "gemini-2-0-flash-thinking-jan-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.5777777777777777,
   "source_value_text": "0.5777777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "gemini-2-0-flash-thinking-jan-2025",
   "benchmark": "aider_polyglot",
   "score": 0.182,
   "source_value_text": "18.2",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-01-21"
  },
  {
   "model": "gemini-2-0-flash-thinking-jan-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.5707070707070707,
   "source_value_text": "0.5707070707070707",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-06"
  },
  {
   "model": "gpt-4-5",
   "benchmark": "math_level_5",
   "score": 0.7862537764350453,
   "source_value_text": "0.7862537764350453",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-28"
  },
  {
   "model": "gpt-4-5",
   "benchmark": "otis_mock_aime",
   "score": 0.37777777777777777,
   "source_value_text": "0.37777777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-28"
  },
  {
   "model": "gpt-4-5",
   "benchmark": "gpqa_diamond",
   "score": 0.6868686868686869,
   "source_value_text": "0.6868686868686869",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-28"
  },
  {
   "model": "gemini-2-0-flash-feb-2025",
   "benchmark": "math_level_5",
   "score": 0.8216578549848943,
   "source_value_text": "0.8216578549848943",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-06"
  },
  {
   "model": "gemini-2-0-flash-feb-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.3111111111111111,
   "source_value_text": "0.3111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-2-0-flash-feb-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.6414141414141414,
   "source_value_text": "0.6414141414141414",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-06"
  },
  {
   "model": "mistral-medium-3",
   "benchmark": "math_level_5",
   "score": 0.8162764350453172,
   "source_value_text": "0.8162764350453172",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-07"
  },
  {
   "model": "mistral-medium-3",
   "benchmark": "otis_mock_aime",
   "score": 0.32222222222222224,
   "source_value_text": "0.32222222222222224",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-07"
  },
  {
   "model": "mistral-medium-3",
   "benchmark": "gpqa_diamond",
   "score": 0.5953282828282829,
   "source_value_text": "0.5953282828282829",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-05-07"
  },
  {
   "model": "mistral-medium-3",
   "benchmark": "bfcl_overall",
   "score": 0.37689999999999996,
   "source_value_text": "37.69%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-4b-instruct-2507",
   "benchmark": "otis_mock_aime",
   "score": 0.5222222222222223,
   "source_value_text": "0.5222222222222223",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-4b-instruct-2507",
   "benchmark": "gpqa_diamond",
   "score": 0.4583333333333333,
   "source_value_text": "0.4583333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-4b-instruct-2507",
   "benchmark": "bfcl_overall",
   "score": 0.3568,
   "source_value_text": "35.68%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-235b-a22b",
   "benchmark": "math_level_5",
   "score": 0.6885699899295065,
   "source_value_text": "0.6885699899295065",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-06-03"
  },
  {
   "model": "qwen3-235b-a22b",
   "benchmark": "aider_polyglot",
   "score": 0.596,
   "source_value_text": "59.6",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-05-09"
  },
  {
   "model": "qwen3-235b-a22b",
   "benchmark": "gpqa_diamond",
   "score": 0.7070707070707071,
   "source_value_text": "0.7070707070707071",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-06-03"
  },
  {
   "model": "deepseek-v3-mar-2025",
   "benchmark": "math_level_5",
   "score": 0.75547583081571,
   "source_value_text": "0.75547583081571",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "deepseek-v3-mar-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.37777777777777777,
   "source_value_text": "0.37777777777777777",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "deepseek-v3-mar-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.6761363636363636,
   "source_value_text": "0.6761363636363636",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "deepseek-v3-mar-2025",
   "benchmark": "or_gpqa_diamond",
   "score": 0.646465,
   "source_value_text": "0.646465",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-29"
  },
  {
   "model": "deepseek-v3-mar-2025",
   "benchmark": "or_tau2_airline",
   "score": 0.4337431393442623,
   "source_value_text": "0.4337431393442623",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-22"
  },
  {
   "model": "qwen3-coder-480b-a35b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.6035354999999999,
   "source_value_text": "0.6035354999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "qwen3-coder-480b-a35b",
   "benchmark": "or_tau2_airline",
   "score": 0.43333342857142854,
   "source_value_text": "0.43333342857142854",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-03"
  },
  {
   "model": "o1-preview",
   "benchmark": "math_level_5",
   "score": 0.8164652567975831,
   "source_value_text": "0.8164652567975831",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "o1-preview",
   "benchmark": "otis_mock_aime",
   "score": 0.3111111111111111,
   "source_value_text": "0.3111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "o1-preview",
   "benchmark": "gpqa_diamond",
   "score": 0.5031565656565656,
   "source_value_text": "0.5031565656565656",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "magistral-small-1-0",
   "benchmark": "otis_mock_aime",
   "score": 0.3,
   "source_value_text": "0.3",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "magistral-small-1-0",
   "benchmark": "gpqa_diamond",
   "score": 0.5606060606060606,
   "source_value_text": "0.5606060606060606",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-06"
  },
  {
   "model": "mistral-small-3-2",
   "benchmark": "otis_mock_aime",
   "score": 0.30277777777777776,
   "source_value_text": "0.30277777777777776",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "mistral-small-3-2",
   "benchmark": "gpqa_diamond",
   "score": 0.49053030303030304,
   "source_value_text": "0.49053030303030304",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "mistral-small-3-2",
   "benchmark": "bfcl_overall",
   "score": 0.3715,
   "source_value_text": "37.15%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gemini-1-5-pro-sept-2024",
   "benchmark": "math_level_5",
   "score": 0.7039274924471299,
   "source_value_text": "0.7039274924471299",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-5-pro-sept-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.23055555555555557,
   "source_value_text": "0.23055555555555557",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-1-5-pro-sept-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.5722853535353535,
   "source_value_text": "0.5722853535353535",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-2-0-flash-dec-2024",
   "benchmark": "aider_polyglot",
   "score": 0.222,
   "source_value_text": "22.2",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-22"
  },
  {
   "model": "magistral-small-1-2",
   "benchmark": "otis_mock_aime",
   "score": 0.28055555555555556,
   "source_value_text": "0.28055555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "magistral-small-1-2",
   "benchmark": "gpqa_diamond",
   "score": 0.476010101010101,
   "source_value_text": "0.476010101010101",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen2-5-max",
   "benchmark": "math_level_5",
   "score": 0.6718277945619335,
   "source_value_text": "0.6718277945619335",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "qwen2-5-max",
   "benchmark": "otis_mock_aime",
   "score": 0.16111111111111112,
   "source_value_text": "0.16111111111111112",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "qwen2-5-max",
   "benchmark": "aider_polyglot",
   "score": 0.218,
   "source_value_text": "21.8",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-01-28"
  },
  {
   "model": "qwen2-5-max",
   "benchmark": "gpqa_diamond",
   "score": 0.5612373737373737,
   "source_value_text": "0.5612373737373737",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "deepseek-v3",
   "benchmark": "math_level_5",
   "score": 0.6485083081570997,
   "source_value_text": "0.6485083081570997",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "deepseek-v3",
   "benchmark": "otis_mock_aime",
   "score": 0.15833333333333333,
   "source_value_text": "0.15833333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "deepseek-v3",
   "benchmark": "gpqa_diamond",
   "score": 0.5653409090909091,
   "source_value_text": "0.5653409090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "deepseek-v3",
   "benchmark": "or_gpqa_diamond",
   "score": 0.661616,
   "source_value_text": "0.661616",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "deepseek-v3",
   "benchmark": "or_tau2_airline",
   "score": 0.4053599698492462,
   "source_value_text": "0.4053599698492462",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-02"
  },
  {
   "model": "glm-4-7-flash-none",
   "benchmark": "otis_mock_aime",
   "score": 0.25,
   "source_value_text": "0.25",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "glm-4-7-flash-none",
   "benchmark": "gpqa_diamond",
   "score": 0.4513888888888889,
   "source_value_text": "0.4513888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen3-4b",
   "benchmark": "gpqa_diamond",
   "score": 0.5233585858585859,
   "source_value_text": "0.5233585858585859",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "phi-4",
   "benchmark": "math_level_5",
   "score": 0.6493580060422961,
   "source_value_text": "0.6493580060422961",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-31"
  },
  {
   "model": "phi-4",
   "benchmark": "otis_mock_aime",
   "score": 0.1375,
   "source_value_text": "0.1375",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "phi-4",
   "benchmark": "gpqa_diamond",
   "score": 0.5606060606060606,
   "source_value_text": "0.5606060606060606",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-31"
  },
  {
   "model": "phi-4",
   "benchmark": "bfcl_overall",
   "score": 0.2879,
   "source_value_text": "28.79%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen-plus-jan-2025",
   "benchmark": "math_level_5",
   "score": 0.6527567975830816,
   "source_value_text": "0.6527567975830816",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-07"
  },
  {
   "model": "qwen-plus-jan-2025",
   "benchmark": "otis_mock_aime",
   "score": 0.17777777777777778,
   "source_value_text": "0.17777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-07"
  },
  {
   "model": "qwen-plus-jan-2025",
   "benchmark": "gpqa_diamond",
   "score": 0.4810606060606061,
   "source_value_text": "0.4810606060606061",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-07"
  },
  {
   "model": "claude-3-5-sonnet-october-2024",
   "benchmark": "math_level_5",
   "score": 0.5694864048338368,
   "source_value_text": "0.5694864048338368",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-5-sonnet-october-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.08472222222222223,
   "source_value_text": "0.08472222222222223",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "claude-3-5-sonnet-october-2024",
   "benchmark": "aider_polyglot",
   "score": 0.516,
   "source_value_text": "51.6",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-01-17"
  },
  {
   "model": "claude-3-5-sonnet-october-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.553030303030303,
   "source_value_text": "0.553030303030303",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "grok-2-dec-2024",
   "benchmark": "math_level_5",
   "score": 0.6351963746223565,
   "source_value_text": "0.6351963746223565",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-17"
  },
  {
   "model": "grok-2-dec-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.11527777777777778,
   "source_value_text": "0.11527777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "grok-2-dec-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.5378787878787878,
   "source_value_text": "0.5378787878787878",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "qwen2-5-coder-32b-instruct",
   "benchmark": "aider_polyglot",
   "score": 0.16399999999999998,
   "source_value_text": "16.4",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-26"
  },
  {
   "model": "gemini-1-5-flash-sep-2024",
   "benchmark": "math_level_5",
   "score": 0.6186744712990937,
   "source_value_text": "0.6186744712990937",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-5-flash-sep-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.1625,
   "source_value_text": "0.1625",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-1-5-flash-sep-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.47316919191919193,
   "source_value_text": "0.47316919191919193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-4-scout",
   "benchmark": "math_level_5",
   "score": 0.6227341389728097,
   "source_value_text": "0.6227341389728097",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-08"
  },
  {
   "model": "llama-4-scout",
   "benchmark": "otis_mock_aime",
   "score": 0.07777777777777778,
   "source_value_text": "0.07777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-08"
  },
  {
   "model": "llama-4-scout",
   "benchmark": "gpqa_diamond",
   "score": 0.5183080808080808,
   "source_value_text": "0.5183080808080808",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-08"
  },
  {
   "model": "llama-4-scout",
   "benchmark": "bfcl_overall",
   "score": 0.2813,
   "source_value_text": "28.13%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gemma-3-12b",
   "benchmark": "otis_mock_aime",
   "score": 0.16666666666666666,
   "source_value_text": "0.16666666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-3-12b",
   "benchmark": "gpqa_diamond",
   "score": 0.39457070707070707,
   "source_value_text": "0.39457070707070707",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-3-12b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.37878799999999996,
   "source_value_text": "0.37878799999999996",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-07"
  },
  {
   "model": "gemma-3-12b",
   "benchmark": "bfcl_overall",
   "score": 0.3043,
   "source_value_text": "30.43%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gemma-3-12b",
   "benchmark": "or_tau2_airline",
   "score": 0.48,
   "source_value_text": "0.48",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-08"
  },
  {
   "model": "deepseek-r1-distill-qwen-1-5b",
   "benchmark": "otis_mock_aime",
   "score": 0.21388888888888888,
   "source_value_text": "0.21388888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "deepseek-r1-distill-qwen-1-5b",
   "benchmark": "gpqa_diamond",
   "score": 0.33585858585858586,
   "source_value_text": "0.33585858585858586",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen2-5-72b",
   "benchmark": "math_level_5",
   "score": 0.631703172205438,
   "source_value_text": "0.631703172205438",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "qwen2-5-72b",
   "benchmark": "otis_mock_aime",
   "score": 0.08055555555555556,
   "source_value_text": "0.08055555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "qwen2-5-72b",
   "benchmark": "gpqa_diamond",
   "score": 0.4914772727272727,
   "source_value_text": "0.4914772727272727",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "qwen2-5-72b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.461279,
   "source_value_text": "0.461279",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-01"
  },
  {
   "model": "qwen2-5-72b",
   "benchmark": "or_tau2_airline",
   "score": 0.25,
   "source_value_text": "0.25",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-09-01"
  },
  {
   "model": "mistral-large-2-nov-2024",
   "benchmark": "math_level_5",
   "score": 0.5028323262839879,
   "source_value_text": "0.5028323262839879",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "mistral-large-2-nov-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.07777777777777778,
   "source_value_text": "0.07777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "mistral-large-2-nov-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.5132575757575758,
   "source_value_text": "0.5132575757575758",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "mistral-large-2-nov-2024",
   "benchmark": "bfcl_overall",
   "score": 0.3837,
   "source_value_text": "38.37%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "yi-lightning",
   "benchmark": "aider_polyglot",
   "score": 0.129,
   "source_value_text": "12.9",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-23"
  },
  {
   "model": "claude-3-5-sonnet",
   "benchmark": "math_level_5",
   "score": 0.5168051359516617,
   "source_value_text": "0.5168051359516617",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-5-sonnet",
   "benchmark": "otis_mock_aime",
   "score": 0.06527777777777778,
   "source_value_text": "0.06527777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "claude-3-5-sonnet",
   "benchmark": "gpqa_diamond",
   "score": 0.5404040404040404,
   "source_value_text": "0.5404040404040404",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "qwen2-5-32b",
   "benchmark": "math_level_5",
   "score": 0.5607061933534743,
   "source_value_text": "0.5607061933534743",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-30"
  },
  {
   "model": "qwen2-5-32b",
   "benchmark": "otis_mock_aime",
   "score": 0.07361111111111111,
   "source_value_text": "0.07361111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "qwen2-5-32b",
   "benchmark": "gpqa_diamond",
   "score": 0.46085858585858586,
   "source_value_text": "0.46085858585858586",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-30"
  },
  {
   "model": "gpt-4o-may-2024",
   "benchmark": "math_level_5",
   "score": 0.5104796072507553,
   "source_value_text": "0.5104796072507553",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4o-may-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.0625,
   "source_value_text": "0.0625",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gpt-4o-may-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.4889520202020202,
   "source_value_text": "0.4889520202020202",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4o-may-2024",
   "benchmark": "or_gpqa_diamond",
   "score": 0.542088,
   "source_value_text": "0.542088",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "gpt-4o-may-2024",
   "benchmark": "or_tau2_airline",
   "score": 0.44,
   "source_value_text": "0.44",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "llama-3-1-405b",
   "benchmark": "math_level_5",
   "score": 0.49773413897280966,
   "source_value_text": "0.49773413897280966",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-1-405b",
   "benchmark": "otis_mock_aime",
   "score": 0.09722222222222222,
   "source_value_text": "0.09722222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "llama-3-1-405b",
   "benchmark": "gpqa_diamond",
   "score": 0.5091540404040404,
   "source_value_text": "0.5091540404040404",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "qwen-turbo",
   "benchmark": "math_level_5",
   "score": 0.5623111782477341,
   "source_value_text": "0.5623111782477341",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-07"
  },
  {
   "model": "qwen-turbo",
   "benchmark": "otis_mock_aime",
   "score": 0.06111111111111111,
   "source_value_text": "0.06111111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-07"
  },
  {
   "model": "qwen-turbo",
   "benchmark": "gpqa_diamond",
   "score": 0.41792929292929293,
   "source_value_text": "0.41792929292929293",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-07"
  },
  {
   "model": "gpt-4-turbo-apr-2024",
   "benchmark": "frontiermath_t13",
   "score": 0.007017543859649123,
   "source_value_text": "0.007017543859649123",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-4-turbo-apr-2024",
   "benchmark": "math_level_5",
   "score": 0.46733383685800606,
   "source_value_text": "0.46733383685800606",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-27"
  },
  {
   "model": "gpt-4-turbo-apr-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.06666666666666667,
   "source_value_text": "0.06666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-27"
  },
  {
   "model": "gpt-4-turbo-apr-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.4659090909090909,
   "source_value_text": "0.4659090909090909",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-27"
  },
  {
   "model": "mistral-large-2-jul-2024",
   "benchmark": "math_level_5",
   "score": 0.4481684290030212,
   "source_value_text": "0.4481684290030212",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mistral-large-2-jul-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.08472222222222223,
   "source_value_text": "0.08472222222222223",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "mistral-large-2-jul-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.49021464646464646,
   "source_value_text": "0.49021464646464646",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mistral-small-3-1",
   "benchmark": "math_level_5",
   "score": 0.4677114803625378,
   "source_value_text": "0.4677114803625378",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-18"
  },
  {
   "model": "mistral-small-3-1",
   "benchmark": "otis_mock_aime",
   "score": 0.058333333333333334,
   "source_value_text": "0.058333333333333334",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-18"
  },
  {
   "model": "mistral-small-3-1",
   "benchmark": "gpqa_diamond",
   "score": 0.47474747474747475,
   "source_value_text": "0.47474747474747475",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-18"
  },
  {
   "model": "openhands-lm-32b-v0-1",
   "benchmark": "aider_polyglot",
   "score": 0.102,
   "source_value_text": "10.2",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2025-04-19"
  },
  {
   "model": "mistral-small-3",
   "benchmark": "math_level_5",
   "score": 0.4481684290030212,
   "source_value_text": "0.4481684290030212",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-30"
  },
  {
   "model": "mistral-small-3",
   "benchmark": "otis_mock_aime",
   "score": 0.06666666666666667,
   "source_value_text": "0.06666666666666667",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "mistral-small-3",
   "benchmark": "gpqa_diamond",
   "score": 0.47285353535353536,
   "source_value_text": "0.47285353535353536",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "claude-3-5-haiku",
   "benchmark": "math_level_5",
   "score": 0.46355740181268884,
   "source_value_text": "0.46355740181268884",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-12"
  },
  {
   "model": "claude-3-5-haiku",
   "benchmark": "otis_mock_aime",
   "score": 0.043055555555555555,
   "source_value_text": "0.043055555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "claude-3-5-haiku",
   "benchmark": "aider_polyglot",
   "score": 0.28,
   "source_value_text": "28.0",
   "evaluator": "Aider",
   "source": "Aider LLM Leaderboards",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_snapshot": "https://raw.githubusercontent.com/Aider-AI/aider/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml",
   "source_sha256": "85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de",
   "verify_url": "https://aider.chat/docs/leaderboards/",
   "retrieved_at": "2026-10-18T12:17:28Z",
   "eval_date": "2024-12-21"
  },
  {
   "model": "claude-3-5-haiku",
   "benchmark": "gpqa_diamond",
   "score": 0.3813131313131313,
   "source_value_text": "0.3813131313131313",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-12"
  },
  {
   "model": "gemini-1-5-pro-may-2024",
   "benchmark": "math_level_5",
   "score": 0.4074773413897281,
   "source_value_text": "0.4074773413897281",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-5-pro-may-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.06805555555555555,
   "source_value_text": "0.06805555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-1-5-pro-may-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.4586489898989899,
   "source_value_text": "0.4586489898989899",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-opus",
   "benchmark": "chess_puzzles",
   "score": 0.03,
   "source_value_text": "0.03",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#chess_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/chess-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "claude-3-opus",
   "benchmark": "math_level_5",
   "score": 0.37481117824773413,
   "source_value_text": "0.37481117824773413",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-opus",
   "benchmark": "otis_mock_aime",
   "score": 0.04722222222222222,
   "source_value_text": "0.04722222222222222",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "claude-3-opus",
   "benchmark": "gpqa_diamond",
   "score": 0.4715909090909091,
   "source_value_text": "0.4715909090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-opus",
   "benchmark": "simpleqa_verified",
   "score": 0.126,
   "source_value_text": "0.126",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#simpleqa_verified.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/simple-qa-verified",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-31"
  },
  {
   "model": "qwen3-1-7b",
   "benchmark": "otis_mock_aime",
   "score": 0.08055555555555556,
   "source_value_text": "0.08055555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-1-7b",
   "benchmark": "gpqa_diamond",
   "score": 0.380050505050505,
   "source_value_text": "0.380050505050505",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "qwen3-1-7b",
   "benchmark": "bfcl_overall",
   "score": 0.2841,
   "source_value_text": "28.41%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen3-0-6b",
   "benchmark": "bfcl_overall",
   "score": 0.23929999999999998,
   "source_value_text": "23.93%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "qwen2-72b",
   "benchmark": "math_level_5",
   "score": 0.39067220543806647,
   "source_value_text": "0.39067220543806647",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "qwen2-72b",
   "benchmark": "gpqa_diamond",
   "score": 0.4078282828282828,
   "source_value_text": "0.4078282828282828",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-1-70b",
   "benchmark": "math_level_5",
   "score": 0.36678625377643503,
   "source_value_text": "0.36678625377643503",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-1-70b",
   "benchmark": "otis_mock_aime",
   "score": 0.03611111111111111,
   "source_value_text": "0.03611111111111111",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "llama-3-1-70b",
   "benchmark": "gpqa_diamond",
   "score": 0.44191919191919193,
   "source_value_text": "0.44191919191919193",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemma-3-4b",
   "benchmark": "otis_mock_aime",
   "score": 0.075,
   "source_value_text": "0.075",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-3-4b",
   "benchmark": "gpqa_diamond",
   "score": 0.23232323232323232,
   "source_value_text": "0.23232323232323232",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-3-4b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.271044,
   "source_value_text": "0.271044",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-15"
  },
  {
   "model": "gemma-3-4b",
   "benchmark": "bfcl_overall",
   "score": 0.1962,
   "source_value_text": "19.62%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "llama-3-2-3b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.17340049999999999,
   "source_value_text": "0.17340049999999999",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-10-13"
  },
  {
   "model": "llama-3-2-3b",
   "benchmark": "bfcl_overall",
   "score": 0.2195,
   "source_value_text": "21.95%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "wizardlm-2-8x22b",
   "benchmark": "math_level_5",
   "score": 0.25736404833836857,
   "source_value_text": "0.25736404833836857",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "wizardlm-2-8x22b",
   "benchmark": "gpqa_diamond",
   "score": 0.43434343434343436,
   "source_value_text": "0.43434343434343436",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-5-flash-may-2024",
   "benchmark": "math_level_5",
   "score": 0.2512273413897281,
   "source_value_text": "0.2512273413897281",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-5-flash-may-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.03888888888888889,
   "source_value_text": "0.03888888888888889",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-1-5-flash-may-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.40372474747474746,
   "source_value_text": "0.40372474747474746",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemma-2-27b",
   "benchmark": "math_level_5",
   "score": 0.2788897280966767,
   "source_value_text": "0.2788897280966767",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemma-2-27b",
   "benchmark": "otis_mock_aime",
   "score": 0.013888888888888888,
   "source_value_text": "0.013888888888888888",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "gemma-2-27b",
   "benchmark": "gpqa_diamond",
   "score": 0.3648989898989899,
   "source_value_text": "0.3648989898989899",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4-jun-2023",
   "benchmark": "mystery_game_puzzles",
   "score": 0.12,
   "source_value_text": "0.12",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gpt-4-jun-2023",
   "benchmark": "math_level_5",
   "score": 0.22970166163141995,
   "source_value_text": "0.22970166163141995",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-4-jun-2023",
   "benchmark": "otis_mock_aime",
   "score": 0.011111111111111112,
   "source_value_text": "0.011111111111111112",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-23"
  },
  {
   "model": "gpt-4-jun-2023",
   "benchmark": "gpqa_diamond",
   "score": 0.30650252525252525,
   "source_value_text": "0.30650252525252525",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-5-flash-8b",
   "benchmark": "otis_mock_aime",
   "score": 0.04583333333333333,
   "source_value_text": "0.04583333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-1-5-flash-8b",
   "benchmark": "gpqa_diamond",
   "score": 0.32954545454545453,
   "source_value_text": "0.32954545454545453",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-12"
  },
  {
   "model": "llama-3-70b",
   "benchmark": "math_level_5",
   "score": 0.225547583081571,
   "source_value_text": "0.225547583081571",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-70b",
   "benchmark": "otis_mock_aime",
   "score": 0.043055555555555555,
   "source_value_text": "0.043055555555555555",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "llama-3-70b",
   "benchmark": "gpqa_diamond",
   "score": 0.40561868686868685,
   "source_value_text": "0.40561868686868685",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mistral-large",
   "benchmark": "math_level_5",
   "score": 0.24461858006042297,
   "source_value_text": "0.24461858006042297",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mistral-large",
   "benchmark": "otis_mock_aime",
   "score": 0.019444444444444445,
   "source_value_text": "0.019444444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "mistral-large",
   "benchmark": "gpqa_diamond",
   "score": 0.38762626262626265,
   "source_value_text": "0.38762626262626265",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "granite-4-0-1b",
   "benchmark": "otis_mock_aime",
   "score": 0.008333333333333333,
   "source_value_text": "0.008333333333333333",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "granite-4-0-1b",
   "benchmark": "gpqa_diamond",
   "score": 0.2398989898989899,
   "source_value_text": "0.2398989898989899",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "hermes-2-theta-llama-3-70b",
   "benchmark": "math_level_5",
   "score": 0.22686933534743203,
   "source_value_text": "0.22686933534743203",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "hermes-2-theta-llama-3-70b",
   "benchmark": "otis_mock_aime",
   "score": 0.025,
   "source_value_text": "0.025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-04-01"
  },
  {
   "model": "hermes-2-theta-llama-3-70b",
   "benchmark": "gpqa_diamond",
   "score": 0.3746843434343434,
   "source_value_text": "0.3746843434343434",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mixtral-8x22b",
   "benchmark": "math_level_5",
   "score": 0.24244712990936557,
   "source_value_text": "0.24244712990936557",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mixtral-8x22b",
   "benchmark": "gpqa_diamond",
   "score": 0.34059343434343436,
   "source_value_text": "0.34059343434343436",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-1-8b",
   "benchmark": "math_level_5",
   "score": 0.22875755287009064,
   "source_value_text": "0.22875755287009064",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-1-8b",
   "benchmark": "otis_mock_aime",
   "score": 0.016666666666666666,
   "source_value_text": "0.016666666666666666",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "llama-3-1-8b",
   "benchmark": "gpqa_diamond",
   "score": 0.26957070707070707,
   "source_value_text": "0.26957070707070707",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "llama-3-1-8b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.26941151269035535,
   "source_value_text": "0.26941151269035535",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-10-17"
  },
  {
   "model": "mistral-7b-v0-3",
   "benchmark": "otis_mock_aime",
   "score": 0.002777777777777778,
   "source_value_text": "0.002777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "mistral-7b-v0-3",
   "benchmark": "gpqa_diamond",
   "score": 0.1518308080808081,
   "source_value_text": "0.1518308080808081",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "granite-4-0-micro",
   "benchmark": "otis_mock_aime",
   "score": 0.027777777777777776,
   "source_value_text": "0.027777777777777776",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "granite-4-0-micro",
   "benchmark": "gpqa_diamond",
   "score": 0.2828282828282828,
   "source_value_text": "0.2828282828282828",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "claude-3-sonnet",
   "benchmark": "math_level_5",
   "score": 0.18174093655589124,
   "source_value_text": "0.18174093655589124",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-sonnet",
   "benchmark": "otis_mock_aime",
   "score": 0.025,
   "source_value_text": "0.025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "claude-3-sonnet",
   "benchmark": "gpqa_diamond",
   "score": 0.4059343434343434,
   "source_value_text": "0.4059343434343434",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemma-2-9b",
   "benchmark": "math_level_5",
   "score": 0.2100641993957704,
   "source_value_text": "0.2100641993957704",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemma-2-9b",
   "benchmark": "otis_mock_aime",
   "score": 0.005555555555555556,
   "source_value_text": "0.005555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "gemma-2-9b",
   "benchmark": "gpqa_diamond",
   "score": 0.2746212121212121,
   "source_value_text": "0.2746212121212121",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "granite-4-0-350m",
   "benchmark": "otis_mock_aime",
   "score": 0.002777777777777778,
   "source_value_text": "0.002777777777777778",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "granite-4-0-350m",
   "benchmark": "gpqa_diamond",
   "score": 0.11237373737373738,
   "source_value_text": "0.11237373737373738",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "granite-4-0-350m",
   "benchmark": "bfcl_overall",
   "score": 0.1898,
   "source_value_text": "18.98%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gpt-4-mar-2023",
   "benchmark": "otis_mock_aime",
   "score": 0.005555555555555556,
   "source_value_text": "0.005555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-22"
  },
  {
   "model": "gpt-4-mar-2023",
   "benchmark": "gpqa_diamond",
   "score": 0.3573232323232323,
   "source_value_text": "0.3573232323232323",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-10-23"
  },
  {
   "model": "claude-2-1",
   "benchmark": "otis_mock_aime",
   "score": 0.019444444444444445,
   "source_value_text": "0.019444444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "claude-2-1",
   "benchmark": "gpqa_diamond",
   "score": 0.32954545454545453,
   "source_value_text": "0.32954545454545453",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "eurus-2-7b-prime",
   "benchmark": "gpqa_diamond",
   "score": 0.3390151515151515,
   "source_value_text": "0.3390151515151515",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "phi-3-medium-14b",
   "benchmark": "math_level_5",
   "score": 0.17560422960725075,
   "source_value_text": "0.17560422960725075",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-31"
  },
  {
   "model": "phi-3-medium-14b",
   "benchmark": "gpqa_diamond",
   "score": 0.2758838383838384,
   "source_value_text": "0.2758838383838384",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-31"
  },
  {
   "model": "qwen2-5-7b",
   "benchmark": "otis_mock_aime",
   "score": 0.025,
   "source_value_text": "0.025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen2-5-7b",
   "benchmark": "gpqa_diamond",
   "score": 0.3547979797979798,
   "source_value_text": "0.3547979797979798",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "qwen2-5-7b",
   "benchmark": "or_gpqa_diamond",
   "score": 0.3417510000000001,
   "source_value_text": "0.3417510000000001",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T13:21:32Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "qwen2-5-7b",
   "benchmark": "or_tau2_airline",
   "score": 0.166667,
   "source_value_text": "0.166667",
   "evaluator": "OpenRouter",
   "source": "OpenRouter benchmark datasets",
   "source_licence": "CC BY 4.0",
   "source_url": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_snapshot": "https://openrouter.ai/api/v1/datasets/exports/benchmark-leaderboard/2026-10-18.csv",
   "source_sha256": "aa7b4c661b02ff153e48bf6cff39c6f09461b4b3bd56a680df9ef8a112128ee2",
   "verify_url": "https://openrouter.ai/benchmarks/tau2-bench-airline",
   "retrieved_at": "2026-10-18T12:17:25Z",
   "eval_date": "2026-08-01"
  },
  {
   "model": "claude-3-haiku",
   "benchmark": "math_level_5",
   "score": 0.1487915407854985,
   "source_value_text": "0.1487915407854985",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-3-haiku",
   "benchmark": "otis_mock_aime",
   "score": 0.018055555555555554,
   "source_value_text": "0.018055555555555554",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "claude-3-haiku",
   "benchmark": "gpqa_diamond",
   "score": 0.3630050505050505,
   "source_value_text": "0.3630050505050505",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-3-5-turbo-nov-2023",
   "benchmark": "math_level_5",
   "score": 0.15889350453172205,
   "source_value_text": "0.15889350453172205",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-3-5-turbo-nov-2023",
   "benchmark": "gpqa_diamond",
   "score": 0.2803030303030303,
   "source_value_text": "0.2803030303030303",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "ministral-3b",
   "benchmark": "math_level_5",
   "score": 0.14444864048338368,
   "source_value_text": "0.14444864048338368",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "ministral-3b",
   "benchmark": "gpqa_diamond",
   "score": 0.25252525252525254,
   "source_value_text": "0.25252525252525254",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-2",
   "benchmark": "math_level_5",
   "score": 0.1172583081570997,
   "source_value_text": "0.1172583081570997",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "claude-2",
   "benchmark": "otis_mock_aime",
   "score": 0.025,
   "source_value_text": "0.025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-03-07"
  },
  {
   "model": "claude-2",
   "benchmark": "gpqa_diamond",
   "score": 0.3465909090909091,
   "source_value_text": "0.3465909090909091",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-3-5-turbo-jan-2024",
   "benchmark": "mystery_game_puzzles",
   "score": 0.03,
   "source_value_text": "0.03",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#mystery_game_puzzles.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/mystery-game-puzzles",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-3-5-turbo-jan-2024",
   "benchmark": "frontiermath_t13",
   "score": 0.0,
   "source_value_text": "0.0",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#frontiermath_tiers_1_3_v2.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-27"
  },
  {
   "model": "gpt-3-5-turbo-jan-2024",
   "benchmark": "math_level_5",
   "score": 0.1163141993957704,
   "source_value_text": "0.1163141993957704",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gpt-3-5-turbo-jan-2024",
   "benchmark": "otis_mock_aime",
   "score": 0.022222222222222223,
   "source_value_text": "0.022222222222222223",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-07"
  },
  {
   "model": "gpt-3-5-turbo-jan-2024",
   "benchmark": "gpqa_diamond",
   "score": 0.27178030303030304,
   "source_value_text": "0.27178030303030304",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "ministral-8b",
   "benchmark": "math_level_5",
   "score": 0.14935800604229607,
   "source_value_text": "0.14935800604229607",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "ministral-8b",
   "benchmark": "gpqa_diamond",
   "score": 0.27146464646464646,
   "source_value_text": "0.27146464646464646",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "ministral-8b",
   "benchmark": "bfcl_overall",
   "score": 0.111,
   "source_value_text": "11.10%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "mistral-nemo",
   "benchmark": "math_level_5",
   "score": 0.1082892749244713,
   "source_value_text": "0.1082892749244713",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mistral-nemo",
   "benchmark": "gpqa_diamond",
   "score": 0.2989267676767677,
   "source_value_text": "0.2989267676767677",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mistral-nemo",
   "benchmark": "bfcl_overall",
   "score": 0.2763,
   "source_value_text": "27.63%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "gemini-1-0-pro",
   "benchmark": "math_level_5",
   "score": 0.11244335347432025,
   "source_value_text": "0.11244335347432025",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemini-1-0-pro",
   "benchmark": "otis_mock_aime",
   "score": 0.011111111111111112,
   "source_value_text": "0.011111111111111112",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-02-25"
  },
  {
   "model": "gemini-1-0-pro",
   "benchmark": "gpqa_diamond",
   "score": 0.33964646464646464,
   "source_value_text": "0.33964646464646464",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "dbrx",
   "benchmark": "math_level_5",
   "score": 0.11650302114803625,
   "source_value_text": "0.11650302114803625",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "dbrx",
   "benchmark": "gpqa_diamond",
   "score": 0.32891414141414144,
   "source_value_text": "0.32891414141414144",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mixtral-8x7b",
   "benchmark": "math_level_5",
   "score": 0.09950906344410876,
   "source_value_text": "0.09950906344410876",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "mixtral-8x7b",
   "benchmark": "gpqa_diamond",
   "score": 0.3058712121212121,
   "source_value_text": "0.3058712121212121",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-2-1b",
   "benchmark": "otis_mock_aime",
   "score": 0.005555555555555556,
   "source_value_text": "0.005555555555555556",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "llama-3-2-1b",
   "benchmark": "gpqa_diamond",
   "score": 0.23926767676767677,
   "source_value_text": "0.23926767676767677",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "llama-3-2-1b",
   "benchmark": "bfcl_overall",
   "score": 0.1082,
   "source_value_text": "10.82%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  },
  {
   "model": "llama-3-8b",
   "benchmark": "math_level_5",
   "score": 0.0612726586102719,
   "source_value_text": "0.0612726586102719",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#math_level_5.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/math-level-5",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "llama-3-8b",
   "benchmark": "otis_mock_aime",
   "score": 0.019444444444444445,
   "source_value_text": "0.019444444444444445",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-30"
  },
  {
   "model": "llama-3-8b",
   "benchmark": "gpqa_diamond",
   "score": 0.26073232323232326,
   "source_value_text": "0.26073232323232326",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2025-01-27"
  },
  {
   "model": "gemma-3-1b",
   "benchmark": "otis_mock_aime",
   "score": 0.011111111111111112,
   "source_value_text": "0.011111111111111112",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#otis_mock_aime_2024_2025.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/otis-mock-aime-2024-2025",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-3-1b",
   "benchmark": "gpqa_diamond",
   "score": 0.1994949494949495,
   "source_value_text": "0.1994949494949495",
   "evaluator": "Epoch AI",
   "source": "Epoch AI Benchmarking Hub",
   "source_licence": "CC BY 4.0",
   "source_url": "https://epoch.ai/data/benchmark_data.zip",
   "source_snapshot": "https://epoch.ai/data/benchmark_data.zip#gpqa_diamond.csv@sha256:e8797a71c56eb5dc",
   "source_sha256": "e8797a71c56eb5dcb04037f3c03711e0a144619adde407df62d65bb2acf2fee6",
   "verify_url": "https://epoch.ai/benchmarks/gpqa-diamond",
   "retrieved_at": "2026-10-18T12:17:21Z",
   "eval_date": "2026-08-28"
  },
  {
   "model": "gemma-3-1b",
   "benchmark": "bfcl_overall",
   "score": 0.0717,
   "source_value_text": "7.17%",
   "evaluator": "UC Berkeley (Gorilla)",
   "source": "Berkeley Function Calling Leaderboard",
   "source_licence": "Apache-2.0",
   "source_url": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_snapshot": "https://raw.githubusercontent.com/ShishirPatil/gorilla/927b29cc0f96061898a3e568ceccbc8bc81ee6a4/data_overall.csv",
   "source_sha256": "ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6",
   "verify_url": "https://gorilla.cs.berkeley.edu/leaderboard.html",
   "retrieved_at": "2026-10-18T12:17:30Z",
   "eval_date": ""
  }
 ]
}