Automated page. Written and translated automatically by AI and checked against the sources by software; no human editor. SI-IQ is an index built from public benchmark results, not an IQ test. How it works

About IQ of SI

IQ of SI answers a simple question — how smart is each AI model? — with a number that anyone can check.

SI-IQ combines published results from independent AI benchmarks into one score per model, on a familiar IQ-style scale where 100 is the average of a fixed reference group of AI models released in 2025. It is an index built from benchmark data, not an IQ test. The methodology gives every formula and parameter, and /data/index.json contains everything needed to recompute every number on this site.

What makes it different

Independent results only. Scores that developers report about their own models are not used in the index.
Openly licensed data. We only build the index from results we are allowed to republish, so the full input data can be downloaded.
A stable scale. The reference group that defines 100 is frozen, so a new model entering the leaderboard does not change existing scores. Method changes are versioned and documented.
Stated uncertainty. Every score has a 90% interval (with its assumptions spelled out on the method page) and models with too little data are not ranked.
Six languages. English, Simplified Chinese, Japanese, Korean, Spanish and German.

How the site is made

The site is produced automatically. A program collects results from the sources every hour, checks them, computes SI-IQ and publishes the pages. The explanatory text on this site (including this page) was written and translated by AI. Pages are assembled from that text and from the data by templates; no person edits individual pages. Numbers are never written by an AI model: they are copied from the sources by code and each new number is checked against its source row.

Corrections

All scores come from the evaluators listed on each model page, together with the exact name the source uses for the model. If a number looks wrong, please check the linked source first: when the evaluator corrects or removes a result, the change normally reaches this site within about two hours (large changes are held a little longer, see the methodology). Our own parsing is checked automatically against every new source row. Publishers can stop our fetcher with robots.txt (see /bot).

Independence

IQ of SI does not run its own tests, sells nothing, shows no advertising and has no affiliate links. Model and company names are used only to identify the models being compared; this site is not affiliated with or endorsed by any AI developer or benchmark publisher.