EarlyDx: leaderboard
EarlyDx (2026) evaluates open-ended early diagnosis from MIMIC-IV emergency department encounters: from the record available at hospital admission, a model lists free-text diagnoses. The board is micro-averaged F1 in percent against the diagnoses that the early evidence supports, matched by a semantic judge, on 6,975 test encounters with zero-shot prompting; higher is better.
Metric: Primary-track F1 (%): micro-averaged F1 of the predicted free-text diagnoses against the evidence-supported ED-encounter diagnoses (matched by a semantic judge), 6,975 test encounters with evidence clipped at hospital admission; zero-shot prompting; higher is better. Source: arxiv.org. Saturation forecast: Around 2033. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 40 | #49 |
| 2 | GLM-5.2 | 37 | #138 |
| 3 | MedGemma-4B | 32 | #2444 |
| 4 | Claude Opus 4.8 | 28 | #56 |
| 5 | Nemotron 3 Ultra | 25 | #429 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=earlydx · How It Works · Data refreshed daily, snapshot 2026-11-13.