MedXpertQA MM Extended (Mercor held-out)
Health & Medicine
Mercor's held-out Extended twin of MedXpertQA's multimodal split: 51 Mercor medical questions that require reasoning over clinical images and patient details, scored by answer accuracy. The public twin is the 200-question MedXpertQA MM set (medxpertqa-mm-200). 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 |
|---|---|
| Claude Opus 5.5 | 71.2 |
| GPT-6 Sol | 68.0 |
| Gemini 4 Argon | 66.7 |
| Claude Fable 5.1 | 66.0 |
| GPT-5.6 Sol Pro | 64.7 |
| GPT-6.1 Sol | 64.7 |
| Claude Opus 5 | 64.1 |
| Gemini 3.7 Flash | 62.7 |
| GPT-5.4 | 62.1 |
| Gemini 3.1 Pro Preview | 60.8 |
| Gemini 3.8 Flash | 60.8 |
| GPT-5.5 | 60.8 |
| GPT-6 Astra | 60.8 |
| Muse Spark 1.3 | 60.1 |
| Muse Spark 1.1 | 58.2 |
| Muse Spark 1.2 | 58.2 |
| Gemini 3.5 Flash | 57.5 |
| Claude Fable 5 | 56.9 |
| GPT-5.6 Luna | 56.2 |
| Claude Sonnet 5.5 | 55.6 |
| Gemini 3.6 Flash | 54.2 |
| GLM-5.3 Flash | 52.9 |
| Grok 4.7 | 52.0 |
| Opus 4.8 | 51.6 |
| Kimi K2.7 Code | 51.6 |
| Opus 4.7 | 51.0 |
| GPT-5.6 Terra | 51.0 |
| Inkling | 51.0 |
| Claude Haiku 5.5 | 48.4 |
| DeepSeek V4.1 Flash | 48.4 |
| Grok 4.6 | 48.4 |
| Gemini 3.5 Flash-Lite | 47.7 |
| Kimi K3 | 47.7 |
| GPT-6 Luna | 46.4 |
| Qwen3.5 397B A17B | 45.1 |
| Sonnet 5 | 43.8 |
| Sonnet 4.6 | 41.2 |
| Grok 4.5 | 41.2 |
| MiniMax M3 | 34.6 |