MedXpertQA MM
Health & Medicine · 2025-01-30
MedXpertQA MM is the full 2,000-question multimodal split of MedXpertQA, expert-level five-option medical questions that pair clinical images such as radiographs, photographs and charts with structured patient information. It is scored by answer accuracy; the 200-question MedXpertQA MM set used by other evaluators is a separate benchmark. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 81.5 |
| Gemini 3.1 Pro Preview | 81.3 |
| Qwen3.8-Max | 80.4 |
| Muse Spark | 78.4 |
| Gemini 3 Pro Preview | 76.0 |
| GPT-5.2 | 73.3 |
| Opus 4.8 | 71.7 |
| Qwen3.7-Plus (source unspecified) | 71.0 |
| Qwen3.5 397B A17B | 70.0 |
| Qwen3.6 Plus (2026-04-02) | 68.7 |
| o3 | 67.5 |
| Qwen3.5 122B A10B | 67.3 |
| Grok 4.20 | 65.8 |
| Kimi K2.5 | 65.3 |
| Opus 4.5 | 63.6 |
| Qwen3.5-27B | 62.4 |
| Qwen3.5 35B A3B | 61.4 |
| Gemma 4 31B IT | 61.3 |
| Gemini 2.5 Pro | 58.9 |
| Gemma 4 26B A4B IT | 58.1 |
| Qwen3.5 Plus (2026-02-15) | 55.3 |
| Qwen3.5-Omni-Plus | 54.7 |
| Sonnet 4.5 | 54.0 |
| Qwen3.5-9B | 49.9 |
| DiffusionGemma 26B A4B IT | 49.0 |
| Gemma 4 12B IT | 48.7 |
| Gemini 2.5 Flash | 47.4 |
| Qwen3.5-Omni-Flash | 44.8 |
| Qwen3.5-4B | 42.9 |
| Gemini 2.5 Flash-Lite | 35.3 |
| GPT-5 Mini | 34.4 |
| Gemma 3 27B IT | 29.8 |
| Gemma 4 E4B IT | 28.7 |
| Qwen3.5-2B | 26.9 |
| GPT-5 Nano | 26.7 |
| Qwen3.5-0.8B | 25.3 |
| MedGemma 4B IT | 24.4 |
| Gemma 4 E2B IT | 23.5 |
| Gemma 3 4B IT | 22.3 |