Medical Reasoning on NEJM (OOD)
68.3AccuracyPRA (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| PRA (Ours)Policy=Qwen3-4B-Instruct2026.04 | 68.3 | |
| RAG + SCPolicy=Qwen3-4B-Instruct2026.04 | 66.9 | |
| RAGPolicy=Qwen3-4B-Instruct2026.04 | 65.7 | |
| CoT + SCPolicy=Qwen3-4B-Instruct2026.04 | 63.2 | |
| CoTPolicy=Qwen3-4B-Instruct2026.04 | 62.7 | |
| Direct + SCPolicy=Qwen3-4B-Instruct2026.04 | 46.4 | |
| DirectPolicy=Qwen3-4B-Instruct2026.04 | 45.3 |