MedLoCoMo: leaderboard
MedLoCoMo (2026) tests long-context clinical reasoning over multi-admission patient histories: 100 synthetic doctor-patient conversations built from MIMIC-IV records, about 74,500 tokens each, with the full history in context. The score in percent combines judged accuracy on answerable questions with abstention accuracy on unanswerable ones over 17,892 questions; higher is better.
Metric: Score (%) over all 17,892 questions: item-weighted combination of judged accuracy on answerable questions (two per admission) and abstention accuracy on unanswerable ones (one per admission); full patient history in context (100 synthetic multi-admission doctor-patient conversations from MIMIC-IV, 74.5k tokens on average), short free-text answers; answerable items judged correct or not by a Gemini 3 Flash Preview judge, adversarial unanswerable items scored by abstention; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 72.8 | #229 |
| 2 | Qwen 3.5 27B | 68.1 | #359 |
| 3 | Qwen 3.5 9B | 51.8 | #803 |
| 4 | Qwen 3.5 4B | 48.8 | #1175 |
| 5 | Gemma 3 27B | 40.3 | #1631 |
| 6 | Gemma 3 12B | 35.7 | #2000 |
| 7 | MedGemma-4B | 18.4 | #2444 |
| 8 | Gemma 3 4B | 12.6 | #5064 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medlocomo · How It Works · Data refreshed daily, snapshot 2026-11-13.