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

Leaderboard

A model is ranked when its results come from at least 5 benchmarks (the two GPQA Diamond runs count as one) in at least 3 areas, from at least 2 different evaluators, and its versions are released no more than 60 days apart. Intervals show the statistical uncertainty of each estimate; models whose intervals overlap may not really differ. Arena is a human-preference rating from crowd votes, shown for reference; it is not part of SI-IQ. See 170 more models with too little data

How fast is AI getting smarter?

Explore the full history of AI capability

How SI-IQ works, in four steps

1. We collect results from 13 public benchmarks run by independent evaluators (not the model makers themselves).
2. Each score is converted to a common scale that accounts for guessing and for how hard the benchmark is.
3. A simple statistical model puts all models and benchmarks on one ability scale, so a model that was tested on fewer or harder benchmarks is still comparable.
4. The scale is set so that a fixed reference group of AI models released in 2025 averages 100 with a standard deviation of 15 points — the same convention IQ scores use.

Read the full methodology · Download the data

Data sources: 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 (licences and changes).