AI Leaderboard: The Best LLMs Across Benchmarks

Only reviewed, active results from independent evaluators count. Numbers that vendors publish in their own model cards stay out. Within each test series the values are standardized, which puts accuracy rates, Elo ratings and pass rates on one shared scale. Percentages and rates are stretched first, so gaps near 0 and 100% are not compressed.

A one-factor statistical model then estimates one shared strength per model and accounts for how sharply each series separates the models. Area scores move away from that strength only as far as the model's own results in the area support. Where a model has no own results in a core area, its overall strength stands in. The note “estimated without own results in …” makes this visible.

The overall score weights core areas such as coding, reasoning, math and knowledge at 75% and the specialist domains finance, law and health at 25%. A model receives an overall score once it has eight test series from at least two sources and three areas.

The uncertainty comes from 200 recalculations in which the test families are drawn again at random. A model is placed once its uncertainty is small enough. A rank range such as “rank 1 to 3” covers the places the model takes in 90% of these recalculations. The winner is the ranked model with the highest point estimate. The win chance is the share of 200 recalculations with randomly redrawn test families in which the model comes first. A recalculation without the model counts as no win for it. A model is statistically level with the winner when it is ahead of the winner in at least 5% of the recalculations that include both.

The index runs from 0 to 100 and is not a percentage. A score of 80 does not mean a model solves 80% of all tasks. It describes the position relative to a fixed reference group of 29 language models whose average sits at 50.

The full calculation with every weight and raw result is available as a JSON export. For sources, test settings and pricing of a single model, browse the AI model directory.