AgentEHR (ReAct): leaderboard
Clinical prediction over MIMIC-IV electronic health records by an LLM agent that queries the patient record through a 19-tool EHR toolbox and predicts the diagnoses, laboratory tests, microbiology tests, prescriptions, procedures and transfers that follow; AgentEHR (Liao et al., 2026). Mean F1 in percent over the six tasks on the 3,248-case MIMIC-IV-Common subset, with the standard ReAct agent limited to 100 turns; higher is better.
Metric: F1 (%; mean of six clinical prediction tasks on the MIMIC-IV-Common subset, 3,248 cases; standard ReAct agent with the 19-tool EHR toolbox, at most 100 turns). Source: arxiv.org. Saturation forecast: Around October 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.1 Fast | 25.01 | #426 |
| 2 | GPT-5 Mini | 23 | #349 |
| 3 | Qwen 3 Next 80B A3B | 18 | #756 |
| 4 | Qwen 3 30B A3B | 15.87 | #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 · How It Works · Data refreshed daily, snapshot 2026-11-12.