Leaderboard
Compare models
Two to four models side by side: their best run on every benchmark, and, for one benchmark, input by input where they disagree most.
Provider
Models whose test configurations are retired, with their best legacy run on each benchmark. They are not ranked.
| Compare | Model | AverageThe average over the benchmarks a model was run on, each benchmark's metric scaled to 0–100, ± its 95% sampling margin. With a task chosen, the average over that task's benchmarks, for models that ran all of them. | Benchmarks | First run |
|---|
Pick two to four models from the list. A model's own page has a button that starts here.