EHRBench - Treatment: leaderboard
Metric: Accuracy (%) on the treatment-decision questions, EHR-grounded multiple-choice clinical decision questions (4, 5 and 6 options) built from MIMIC-III, MIMIC-IV and PROMOTE encounter trajectories with knowledge-base verification, JSON-constrained answers, deterministic decoding; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 31 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 Chat | 80.45 |
| 2 | GPT-5.2 Instant | 80.13 |
| 3 | GPT-4.1 | 80.1 |
| 4 | Llama 3.3 70B Instruct | 79.05 |
| 5 | GPT-4.1 Mini | 77.9 |
| 6 | GLM-4 32B (0414) | 77.9 |
| 7 | Qwen 3 32B | 77.34 |
| 8 | Qwen 2.5 32B | 76.87 |
| 9 | GPT-5 Mini | 76.17 |
| 10 | Mistral Small 3 | 75.02 |
| 11 | Qwen 3 8B | 74.49 |
| 12 | GPT-4.1 Nano | 74.03 |
| 13 | Qwen 3 4B | 73.46 |
| 14 | GLM-4 9B (0414) | 72.61 |
| 15 | Yi 1.5 34B | 72.25 |
Interactive version: theaggregate.ai/benchmark?slug=ehrbench-treatment · How It Works · Data refreshed daily, snapshot 2026-10-07.