MINT (Medical) - MedBullets (Question Last): leaderboard

MINT (2026) turns medical multiple-choice diagnosis cases into conversations that reveal one clinically labelled evidence shard per turn. In the Ask-Question-Last setting the question and options arrive only after all shards; this board scores the final answer on the MedBullets cases. Accuracy in percent; higher is better.

Metric: Final-answer accuracy (%) on the 62 MedBullets-derived diagnosis cases, each split into evidence shards revealed one per turn, with the multiple-choice diagnostic question shown only at the last turn (Ask-Question-Last); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1O4 Mini83.6#368
2GPT-5 Mini80.3#364
3Claude Sonnet 4.673.8#173
4MINT MedGemma-27B (checkpoint unspecified)66.1
5GPT-OSS-20B64.5#1183
6Med42-v2-70B64.5
7Qwen 3 4B56.5#2348
8Med42-v2-8B54.8#2558
9MedGemma 1.5 4B46.8#2716
10MINT Qwen2.5-3B (checkpoint unspecified)40.3
11MINT OLMo-2-7B (checkpoint unspecified)25.8

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

Interactive version: theaggregate.ai/benchmark?slug=mint-medical-medbullets-question-last · How It Works · Data refreshed daily, snapshot 2026-11-21.