ClinicalMC (English): leaderboard
ClinicalMC (Hou et al., 2026) on 5,804 English multi-course cases rebuilt from PMC-Patients reports: acting as the doctor in a simulated hospital with GPT-4o-mini patient and examiner agents, the model triages, orders examinations, diagnoses and plans treatment at admission, then assesses and treats each later course and gives the discharge diagnosis, each stage given the gold record of the earlier ones. The score is the mean of eleven task scores in percent (accuracy, recall, F1, IoU and LLM-judged reasoning quality), higher is better.
Metric: Mean of 11 task scores (%; triage accuracy, examination recall, diagnosis F1, LLM-judged diagnosis basis and differential diagnosis, treatment and per-course IoU; GPT-4o-mini patient and examiner agents, gold inputs from earlier stages, mean of three runs; 5,804 English cases from PMC-Patients). Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Mixtral 8x22B Instruct | 49.8 | #2052 |
| 2 | Qwen Turbo | 49.68 | #2002 |
| 3 | GPT-5 Mini | 46.88 | #349 |
| 4 | GPT-4o Mini | 46.23 | #1627 |
| 5 | Qwen 2.5 32B Instruct | 45.98 | #1216 |
| 6 | Qwen 3 Next 80B A3B Instruct | 45.59 | #763 |
| 7 | Qwen 2.5 14B Instruct | 45.38 | #1745 |
| 8 | Qwen 2.5 72B Instruct | 45.15 | #1093 |
| 9 | Mistral 7B Instruct (v0.3) | 45.06 | #6698 |
| 10 | Qwen 2.5 7B Instruct | 44.98 | #2583 |
| 11 | Llama 3.3 70B Instruct | 44.52 | #1353 |
| 12 | Falcon3-7B-Instruct | 43.67 | #6543 |
| 13 | Llama 3.2 3B Instruct | 36.47 | #7185 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clinicalmc-english · How It Works · Data refreshed daily, snapshot 2026-11-12.