MC-CXR - Image-Only Recognition: leaderboard
MC-CXR (Seoul National University, 2026) tests whether vision-language models keep a correct chest X-ray reading when auxiliary context is reliable or misleading, over 240 radiologist-reviewed MIMIC-CXR cases with a fixed target finding each. This board is the image-only baseline: accuracy in percent of the single-letter answer among 12 CheXpert labels (chance about 8); higher is better.
Metric: Image-only recognition accuracy (%; the image alone, no context; 240 radiologist-reviewed MIMIC-CXR cases, one current chest X-ray and a fixed target finding each; zero-shot constrained single-letter answer from 12 CheXpert labels (chance about 8.3%), invalid answers scored wrong; greedy or temperature-0 decoding). Source: arxiv.org. Saturation forecast: Around 2032. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MedGemma 1.5 4B | 31.7 | #2562 |
| 2 | Qwen 3.5 9B | 30 | #803 |
| 3 | Gemini 3.5 Flash (Minimal) | 28.3 | #101 (Gemini 3.5 Flash) |
| 4 | InternVL3-8B | 26.7 | #1688 |
| 5 | Claude Opus 4.7 | 21.2 | #79 |
| 6 | Phi-4 Multimodal Instruct | 15.8 | #3882 |
| 7 | Gemma 3 12B (IT) | 9.2 | #1837 |
| 8 | Llama 3.2 11B Instruct | 8.3 | #4991 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mc-cxr-image-only-recognition · How It Works · Data refreshed daily, snapshot 2026-11-13.