ClinTraceBench - Retrospective Propagation: leaderboard
ClinTraceBench (Wang, Zhao and Zhao, Findings of EMNLP 2026) tests longitudinal clinical reasoning over multi-visit patient-clinician dialogues derived from MIMIC-IV records; this task asks for the earlier findings that support a later diagnosis, credited only when the set of event IDs matches exactly. Accuracy in percent over 53 questions with the whole dialogue history in context; higher is better.
Metric: Accuracy (%; task T4, retrospective propagation: list the earlier findings supporting a later diagnosis, scored 1 only for an exact event-ID set, 53 questions; full-context strategy: the whole multi-visit patient-clinician dialogue derived from MIMIC-IV records is given verbatim, models called through OpenRouter). Source: arxiv.org. Saturation forecast: Around 2032. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o Mini | 28.3 | #1627 |
| 2 | Claude Sonnet 4.6 | 24.5 | #158 |
| 3 | DeepSeek V3.1 | 22.6 | #500 |
| 4 | Claude Haiku 4.5 | 18.9 | #568 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clintracebench-retrospective-propagation · How It Works · Data refreshed daily, snapshot 2026-11-12.