Med-R2 - Lesion Level: leaderboard

Accuracy (%) of medical vision-language models on the Med-R2 (2026) lesion level task, detecting and characterizing lesions, with equal numbers of positive, negative and neutral adversarial question forms; higher is better.

Metric: Accuracy (%) on the Med-R2 Lesion Level task (detecting and characterizing lesions), zero-shot with system prompts; each task mixes equal numbers of positive, negative and neutral adversarial question forms; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Hulu-Med-14B68.6#1607
2HuluMed-7B58.8#2049
3Hulu-Med-4B57.9#2392
4Qwen 2.5 VL 7B Instruct55.1#1902
5InternVL3.5-8B54#1604
6HuaTuoGPT-Vision-34B53.5#1828
7Qwen 3 VL 8B52#917
8Qwen 3 VL 235B A22B51.3#485
9HuatuoGPT-Vision-7B50.5#2162
10GPT-5.2Thinking49.9#185
11Janus-Pro-7B49.5#6857
12GPT-4o47#751
13Lingshu-7B41.6#1635
14MedGemma-4B32.5#2557

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=med-r2-lesion-level · How It Works · Data refreshed daily, snapshot 2026-11-21.