Benchmark

ProgramBench

Program-synthesis and implementation benchmark.

Score history →

Model scores

#ModelScore
1Claude Opus 537%
2DeepSeek V4.1-Flash 101729.1%
3GPT 5.6-Sol23%
4DeepSeek V4.1-Flash 091020.3%
5GLM 5.319%
6Kimi K317.5%
7DeepSeek V4-Pro 081315.5%