Every model's score on every benchmark we track, each linked to where it was published. Independent runs and lab-reported numbers are kept apart. Many of these benchmarks feed the Respan Index.
177 of 177 benchmarks
Reasoning
4 models, higher is better
Abstract visual puzzles from the ARC Prize Foundation that people solve easily but AI systems find hard, testing whether a model can learn a new rule from a few examples.
Benchmark data from llmmetric.com, Respan's model data service.