Benchmark
Humanity's Last Exam — a broad frontier-knowledge set. Some models report a text-only subset (marked with an asterisk at the source).
Score history →| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5.5 | 64.4% |
| 2 | Claude Opus 5 | 56.3% |
| 3 | GPT-6 Astra | 54.7% |
| 4 | GPT 5.6-Sol | 44.5% |
| 5 | Gemini 3.1 Pro | 44.4% |
| 6 | Kimi K3 | 43.5% |
| 7 | DeepSeek V4-Pro 0813 | 42.7% |
| 8 | DeepSeek V4.1-Flash 1017 | 42.6% |
| 9 | GLM 5.3 | 42% |
| 10 | DeepSeek V4-Flash 0731 | 37.8% |
| 11 | DeepSeek V4.1-Flash 0910 | 36.8% |