Benchmark

HLE (w/tools)

HLE run with tool access enabled.

Score history →

Model scores

#ModelScore
1Claude Opus 5.567.7%
2DeepSeek V4.1-Flash 091063.9%
3Claude Opus 563.6%
4GLM 5.362.5%
5DeepSeek V4-Pro 081360%
6Kimi K359.8%
7DeepSeek V4-Flash 073151.5%
8Gemini 3.1 Pro51.4%