Medical Reasoning on MedXpertQA
42.9AccuracyDeepSeek-R1
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeek-R1Backbone=DeepSeek-R12026.07 | 42.9 | |
| Claude Sonnet 4.5Backbone=Claude Sonnet 4.52026.07 | 39.4 | |
| MedLABackbone Model=deepseek-r1, Sample Count=602025.09 | 36 | |
| FaithMed (Qwen3 4B)Backbone=Qwen3 4B2026.07 | 28.2 | |
| MedLABackbone Model=deepseek-v3, Sample Count=602025.09 | 25.6 | |
| Agentic Search + GRPO (Qwen3 4B)Backbone=Qwen3 4B2026.07 | 22 | |
| baselineBackbone Model=deepseek-r1, Sample Count=602025.09 | 21.3 | |
| Agentic Search (Qwen3 4B)Backbone=Qwen3 4B2026.07 | 19.6 | |
| Qwen3 8BBackbone=Qwen3 8B2026.07 | 19.2 | |
| MedGemma 27BBackbone=MedGemma 27B2026.07 | 19 | |
| MedRAG (Qwen3 4B)Backbone=Qwen3 4B2026.07 | 19 | |
| FaithMed (Qwen3 1.7B)Backbone=Qwen3 1.7B2026.07 | 17.1 | |
| Gemma 3 12BBackbone=Gemma 3 12B2026.07 | 16.4 | |
| Base (Qwen3 4B)Backbone=Qwen3 4B2026.07 | 16 | |
| baselineBackbone Model=deepseek-v3, Sample Count=602025.09 | 15 | |
| MedRAG (Qwen3 1.7B)Backbone=Qwen3 1.7B2026.07 | 14.2 | |
| Llama 3.1 8B InstructBackbone=Llama 3.1 8B2026.07 | 11.8 | |
| Agentic Search + GRPO (Qwen3 1.7B)Backbone=Qwen3 1.7B2026.07 | 11.5 | |
| Agentic Search (Qwen3 1.7B)Backbone=Qwen3 1.7B2026.07 | 11.1 | |
| Base (Qwen3 1.7B)Backbone=Qwen3 1.7B2026.07 | 9.3 |