Clinical Diagnosis on MedXpertQA (OOD)
22.97Exact Match (EM)DeepSeek-V4-Pro
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DeepSeek-V4-ProModel Type=General-purpose LLM2026.05 | 22.97 | 24.63 | 23.8 | 24.48 | |
| GPT-5.2Model Type=General-purpose LLM2026.05 | 8.12 | 13.31 | 10.71 | 20.25 | |
| C-MIGBackbone=Qwen-2.5-3B2026.05 | 7.33 | 12.08 | 9.71 | 24.72 | |
| AutoRefine-HardSearchBackbone=Qwen-2.5-3B, Approach=Clinical Sparse Reward2026.05 | 6.73 | 12.12 | 9.43 | 21.79 | |
| AutoRefine-ICDTreeBackbone=Qwen-2.5-3B, Approach=Clinical Sparse Reward2026.05 | 6.14 | 13.74 | 9.94 | 22.62 | |
| IGPOBackbone=Qwen-2.5-3B, Approach=RAG+RL2026.05 | 5.54 | 11.01 | 8.28 | 22.2 | |
| AutoRefine-HardDocBackbone=Qwen-2.5-3B, Approach=Clinical Sparse Reward2026.05 | 4.36 | 7.52 | 5.94 | 20.77 | |
| AutoRefineBackbone=Qwen-2.5-3B, Approach=RAG+RL2026.05 | 3.96 | 8.51 | 6.23 | 20.56 | |
| AutoRefine-EmbeddingBackbone=Qwen-2.5-3B, Approach=Clinical Sparse Reward2026.05 | 3.76 | 8.95 | 6.35 | 22.68 | |
| Llama3-Med42-70BModel Type=Medical LLM2026.05 | 3.37 | 7.56 | 5.47 | 11.25 | |
| HuatuoGPT2-13BModel Type=Medical LLM2026.05 | 1.19 | 4.28 | 2.73 | 2.09 | |
| Baichuan2-13B-ChatModel Type=Medical LLM2026.05 | 1.19 | 3.01 | 2.1 | 7.9 | |
| Search-R1Backbone=Qwen-2.5-3B, Approach=RAG+RL2026.05 | 0.99 | 1.82 | 1.41 | 19.6 | |
| Qwen2.5-3BDescription=directly inference without any training2026.05 | 0.79 | 2.57 | 1.68 | 0.84 |