MMRareBench - Treatment Planning: leaderboard

MMRareBench (2026) treatment-planning track: given the working diagnosis, case narrative and medical images of a rare-disease case, the model writes a stage-wise plan covering safety, monitoring and follow-up. Rubric score (0-100) from a strict Qwen3-VL-235B judge; higher is better.

Metric: Treatment Planning track (T2) score: a stage-wise treatment plan given the working diagnosis, case narrative and images, graded 0-100 by a Qwen3-VL-235B judge with a strict track-specific rubric that awards credit only when a criterion is fully met (six binary dimensions averaged); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 23 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-549.2#178
2Qwen 3 VL 235B A22B Instruct35.9#532
3Claude Sonnet 4.535.8#261
4Claude Haiku 4.534.2#583
5Gemini 3 Flash (Preview)29.1#152
6Gemini 2.5 Pro26.7#276
7GLM-4.6V20.5#710
8Qwen 3 VL 30B A3B Instruct20.1#895
9Gemini 2.5 Flash17.1#463
10MedGemma-27B-IT15.4#1350
11Lingshu-32B14.2#1113
12Hulu-Med-14B14.2#1607
13HealthGPT-L14B14.1#1397
14Qwen 2.5 VL 32B Instruct12.8#1062
15GPT-4o12.7#751

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

Interactive version: theaggregate.ai/benchmark?slug=mmrarebench-treatment-planning · How It Works · Data refreshed daily, snapshot 2026-11-21.