MMRareBench - Examination Suggestion: leaderboard

MMRareBench (2026) examination track: from a rare-disease case narrative cut before the decisive tests, plus its images, the model recommends prioritized next-step investigations with rationale. Rubric score (0-100) from a strict Qwen3-VL-235B judge; higher is better.

Metric: Examination Suggestion track (T4) score: prioritized next-step investigations for a case narrative truncated before the gold tests, 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 (eight 0-2 dimensions normalized by 16); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 23 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.585.9#261
2Gemini 2.5 Pro85.5#276
3Qwen 3 VL 235B A22B Instruct83.2#532
4Gemini 3 Flash (Preview)81.8#152
5Claude Haiku 4.580.6#583
6GPT-577.9#178
7Gemini 2.5 Flash66.9#463
8Qwen 3 VL 30B A3B Instruct63.1#895
9MedGemma-27B-IT56#1350
10Qwen 3 VL 8B Instruct43.9#1029
11GPT-4o42.8#751
12Qwen 2.5 VL 72B Instruct42.3#841
13Qwen 2.5 VL 32B Instruct42#1062
14Hulu-Med-14B38.8#1607
15Hulu-Med-4B37.2#2392

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

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