Humanity's Last Exam Diamond (Knowledge Partition)
General QA
The 500-question knowledge partition of Humanity's Last Exam Diamond (the Scale AI and Center for AI Safety 1,000-question revision, split evenly between reasoning and knowledge questions). Accuracy is the share of the partition's questions judged correct, pass@1, from the same run as the overall HLE-Diamond score.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Carbon | 58.6 |
| GPT-6 Astra | 49.6 |
| Gemini 4 Argon | 48.6 |
| GPT-6.1 Sol | 43.6 |
| Claude Sonnet 5.5 | 38.6 |
| Muse Wonder Full | 35.8 |
| GPT-6 Sol | 30.8 |
| Kimi K3.1 Pro | 29.8 |
| GPT-6.1 Terra | 29.2 |
| Gemini 3.8 Flash | 29.0 |
| Gemini 3.1 Pro Preview | 28.0 |
| Muse Wonder | 27.6 |
| Muse Spark 1.3 | 24.4 |
| Step 5 Preview | 22.0 |
| MiniMax M3.1 Flash | 21.6 |
| Kimi K3 | 20.8 |
| Muse Spark Open | 19.6 |
| GPT-5.6 Terra | 17.2 |
| MiniMax M3 | 17.0 |
| Grok 4.7 | 16.0 |
| DeepSeek V4.1 Flash 1017 | 15.8 |
| DeepSeek V4 Flash Vision Exp | 15.0 |
| Claude Haiku 5.5 | 14.6 |
| GLM-5.3 Flash | 14.4 |
| Qwen3.8 Max (0902) | 14.4 |
| DeepSeek V4 Pro 0813 | 13.8 |
| DeepSeek V4.1 Flash | 13.6 |
| GPT-6 Luna | 13.4 |
| GLM-5.3 | 12.4 |
| GPT-5.5 Instant (2026-06-25 hosted snapshot) | 11.4 |
| MiMo-V2.6-Pro | 11.4 |
| Kimi K2.7 Code | 11.2 |
| Ling 3.1 Flash 1019 | 11.2 |
| Mistral Large 4 | 11.0 |
| LongCat-2.0 | 10.4 |
| MiMo-V2.6-Flash | 10.0 |
| Qwen3.7-Plus (source unspecified) | 9.8 |
| Gemini 3.5 Flash-Lite | 9.4 |
| Qwen3.8-Flash-Next | 9.2 |
| Qwen3.5 397B A17B | 9.0 |