MIMIC-CDM: leaderboard
Clinical decision-making over real de-identified MIMIC admissions for four acute abdominal pathologies: the model reads history, labs and imaging reports and has to name the diagnosis. This board is the average over the four pathologies.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 3 27B (IT) | 70.48 | #523 |
| 2 | Gemma 2 27B | 68.65 | #806 |
| 3 | Gemma 3 12B (IT) | 66.2 | #664 |
| 4 | Llama 3 70B Instruct | 63.32 | #634 |
| 5 | Llama 3.3 70B Instruct | 62.42 | #534 |
| 6 | Llama 3.1 70B Instruct | 60.45 | #561 |
| 7 | MedGemma-4B-IT | 53.05 | #853 |
| 8 | DeepSeek R1 Distill Llama 70B | 51.6 | #594 |
| 9 | Gemma 3 4B (IT) | 50.8 | #992 |
| 10 | Gemma 3 1B (IT) | 21.45 | #1504 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mimic-cdm · How It Works · Data refreshed daily, snapshot 2026-10-20.