HumanEval Benchmark

Code generation benchmark. Max score: 100.

ModelScore
o3 95.2
Claude Opus 4.6 94.0
o4 mini 93.4
Claude Sonnet 4.5 93.0