AgentEHR (ReAct) - Procedures: leaderboard
The procedures task of AgentEHR (Liao et al., 2026): from a MIMIC-IV history censored just before the event, a ReAct agent with the 19-tool EHR toolbox predicts the set of procedures recorded next. F1 in percent between the predicted and recorded sets over 529 cases of the MIMIC-IV-Common subset; higher is better.
Metric: F1 (%; predicted procedures label set against the next recorded events, 529 MIMIC-IV-Common cases; standard ReAct agent). Source: arxiv.org. Saturation forecast: Around August 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.1 Fast | 37.62 | #426 |
| 2 | Qwen 3 Next 80B A3B | 28.57 | #756 |
| 3 | GPT-5 Mini | 26.93 | #349 |
| 4 | Qwen 3 30B A3B | 24.72 | #1124 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=agentehr-react-procedures · How It Works · Data refreshed daily, snapshot 2026-11-12.