MC-CXR - Misleading Preliminary Note: leaderboard

Metric: Accuracy under misleading context (%; the image plus a preliminary note rewritten toward a plausible wrong finding; 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

#ModelScore
1MedGemma 1.5 4B12.1
2Claude Opus 4.711.7
3Gemini 3.5 Flash (Minimal)10.4
4Qwen 3.5 9B9.2
5InternVL3-8B6.2
6Gemma 3 12B (IT)4.6
7Phi-4 Multimodal Instruct4.2
8Llama 3.2 11B Instruct3.3

Interactive version: theaggregate.ai/benchmark?slug=mc-cxr-misleading-preliminary-note · How It Works · Data refreshed daily, snapshot 2026-09-29.