Medical Evidence Auditing on DrugAudit (test)
0.632EILlama ScoreDrugClaw
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DrugClawMode=graph2026.05 | 0.632 | 0.623 | 0.685 | 0.918 | 0.852 | |
| DrugClawMode=linear2026.05 | 0.626 | 0.591 | 0.656 | 0.916 | 0.887 | |
| DrugClawMode=graph, web-search fallback=disabled2026.05 | 0.622 | 0.628 | 0.663 | 0.931 | 0.876 | |
| Direct LLMModel=GPT-52026.05 | 0.513 | 0.494 | 0.401 | 0.714 | 0.825 | |
| Biomni2026.05 | 0.489 | 0.47 | 0.356 | 0.793 | 0.787 | |
| DeepEvidence2026.05 | 0.477 | 0.457 | 0.461 | 0.603 | 0.732 | |
| Direct LLMModel=GPT-5-mini2026.05 | 0.43 | 0.403 | 0.268 | 0.817 | 0.815 | |
| ToolUniverse2026.05 | 0.357 | 0.34 | 0.146 | 0.796 | 0.828 |