Frequently asked questions
Which AI is the smartest right now?
On SI-IQ, GPT-6 Astra from OpenAI currently has the highest score: 150 (90% interval 146–153). Its interval overlaps with those of 2 other ranked models, including GPT-6.1 Sol (148), so the data do not clearly separate it from them. Rankings change as new results are published.
What is SI-IQ?
SI-IQ is an index that combines results from independent AI benchmarks into a single number per model. On its scale, a fixed reference group of AI models released in 2025 averages 100, and one standard deviation of that group is 15 points.
Is SI-IQ a real IQ test?
No. It does not test AI models with IQ questions and it cannot be compared with human IQ scores. It only borrows the familiar scale to make differences between AI models easier to read.
Is GPT or Claude smarter?
It depends on the version. Right now the best OpenAI model on SI-IQ is GPT-6 Astra (150) and the best Anthropic model is Claude Opus 5.5 (144). Check the interval: differences inside the uncertainty are not meaningful.
What is the smartest open-weights model?
The highest-ranked model with openly downloadable weights is currently DeepSeek V4.1 Flash from DeepSeek, with an SI-IQ of 128.
How often is it updated?
We check every source every hour. New models appear as soon as an independent evaluator publishes enough results; the statistical model is refitted once a day.
Why are some published scores missing?
Scores that model makers report about their own models are neither used nor shown, because independent evaluations are more comparable. Results whose licence does not allow republishing are not used either.
Why is a new model not ranked yet?
A model needs at least four independent benchmark results from at least two different evaluators to be ranked. Until then it appears with a provisional estimate.