Closed-form Medical Question Answering on PubMedQA (test)
69.3AccuracyDrugClaw
Evaluation results
| Method | Links | |
|---|---|---|
| DrugClawMode=graph2026.05 | 69.3 | |
| DrugClawMode=graph, web-search fallback=disabled2026.05 | 67.8 | |
| Biomni2026.05 | 66.4 | |
| ToolUniverse2026.05 | 66.2 | |
| DrugClawMode=linear2026.05 | 66 | |
| Direct LLMModel=GPT-52026.05 | 65.4 | |
| Direct LLMModel=GPT-5-mini2026.05 | 65.2 | |
| DeepEvidence2026.05 | 65.2 |