CURE (Clinical Diagnosis, RAG) - Multiple Choice: leaderboard
Metric: Differential-diagnosis accuracy in percent choosing the diagnosis among 4 to 5 candidates (terminology standardized and distractors generated where needed by GPT-5.2, physician-checked on 50 cases), on CURE's 500 test cases (Eurorad 2025 radiology cases: clinical history, a mean of 8 images, radiologist findings with diagnostic statements removed), zero-shot, given the clinical history, images and the top PubMed abstracts retrieved by a fixed bge-large-en-v1.5 plus MedCPT pipeline; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 88 | #64 |
| 2 | GPT-5.2 | 85.6 | #105 |
| 3 | Kimi K2.5 | 83.4 | #139 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 79.8 | #228 (Qwen 3 VL 235B A22B) |
| 5 | Qwen 3 VL 235B A22B Instruct | 77.6 | #264 |
| 6 | GLM-4.6V | 77.4 | #309 |
| 7 | Qwen 3 VL 32B Instruct | 76.2 | #276 |
| 8 | Qwen 3 VL 30B A3B Instruct | 75.6 | #365 |
| 9 | Qwen 3 VL 32B (Thinking) | 73.8 | #287 (Qwen 3 VL 32B) |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 73 | #338 (Qwen 3 VL 30B A3B) |
| 11 | Qwen 3 VL 8B Instruct | 69.2 | #401 |
| 12 | Qwen 3 VL 8B (Thinking) | 67.6 | |
| 13 | Gemma 3 12B (IT) | 63.6 | #655 |
| 14 | MedGemma-4B-IT | 58.2 | #842 |
| 15 | Gemma 3 4B (IT) | 47.8 | #971 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=cure-clinical-diagnosis-rag-multiple-choice · How It Works · Data refreshed daily, snapshot 2026-10-11.