MR-Bench - Procedure Selection: leaderboard
Metric: Accuracy (%, times 100) at picking the admission's procedure set against other patients' procedures; 1,000 MIMIC-IV hospital admissions, eight-option questions, temperature 0.5, an answer counted from the first or the last option letter the output names, whichever scores higher over the dataset; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 73.8 |
| 2 | DeepSeek V3.2 (Thinking) | 69.4 |
| 3 | Qwen 3 Max | 68.2 |
| 4 | Gemini 3 Pro | 67.6 |
| 5 | GPT-4o | 64.9 |
| 6 | Qwen 3 8B | 60.2 |
| 7 | Qwen 2.5 7B Instruct | 59.5 |
| 8 | Qwen 3 4B 2507 (Thinking) | 58.8 |
| 9 | Gemma 3 4B (IT) | 50.4 |
| 10 | Llama 3 8B Instruct | 43.1 |
| 11 | Llama 3.1 8B Instruct | 41.6 |
| 12 | MedGemma-4B | 32.4 |
Interactive version: theaggregate.ai/benchmark?slug=mr-bench-procedure-selection · How It Works · Data refreshed daily, snapshot 2026-10-07.