HLE Extended (Mercor held-out)
General QA
Mercor's held-out Extended twin of Humanity's Last Exam: 447 Mercor closed-ended frontier academic questions, scored by answer accuracy. Mercor does not say whether the set includes image questions. The public twin Mercor reruns is the 2,158-question text-only HLE (hle-text-only, under hle). Mercor does not publish the held-out tasks or say how they were drawn or written, so this is its own benchmark, not a component of the public one.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 77.3 |
| GPT-5.6 Terra | 70.7 |
| Claude Opus 5 | 61.0 |
| Claude Fable 5 | 58.7 |
| GPT-5.6 Luna | 57.5 |
| GPT-5.4 | 56.1 |
| GPT-6 Luna | 53.3 |
| Opus 4.8 | 49.4 |
| Grok 4.6 | 45.8 |
| DeepSeek V4.1 Flash | 45.3 |
| Gemini 3.8 Flash | 36.8 |
| Kimi K3 | 35.0 |
| Gemini 3.7 Flash | 31.6 |
| Opus 4.7 | 31.2 |
| Sonnet 5 | 29.1 |
| GLM-5.2 | 25.8 |
| Grok 4.5 | 25.4 |
| Gemini 3.1 Pro Preview | 19.6 |
| Gemini 3.5 Flash | 14.8 |
| Kimi K2.7 Code | 13.7 |
| Gemini 3.6 Flash | 13.3 |
| Sonnet 4.6 | 10.9 |
| DeepSeek-V3.2 | 8.1 |
| gpt-oss-120b | 8.1 |
| MiniMax M2.7 | 6.6 |
| MiniMax M3 | 5.4 |
| Qwen3.5 397B A17B | 4.2 |
| GLM-5 | 2.0 |