MediLongChat - In-Dialogue Reasoning: leaderboard
MediLongChat (Hu et al., South-Central Minzu University and Singapore Management University, 2026) builds synthetic longitudinal doctor-patient dialogue histories for 80 patients from knowledge-guided records, about 50,000 tokens over 18 visits per patient. In-dialogue reasoning asks for facts from a single consultation, such as the visit date, complaint, diagnosis, medication or treatment plan, as a short answer; the score is token-level F1 against the reference answer in percent, higher is better.
Metric: Token F1 (%) of short answers about a single consultation. Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 27.15 | #775 |
| 2 | GPT-4o Mini | 23.3 | #1627 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medilongchat-in-dialogue-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.