HumanEval

Hand-written programming problems scored by unit tests — a classic code-generation benchmark (pass@1).

Code

Code generation Developer assistants

Higher is better · pass@1 %

1 Claude Opus 5 94.1%

2 Gemini 2.5 Pro 92%

3 GPT OSS 120B 89.5%

4 DeepSeek-V4 Flash 88.4%

Bars scaled from 80 so close scores stay legible — the figure at right is the actual pass@1 %.