DrugAudit: leaderboard
Drug-information question answering with provenance (Wang et al., 2026): 3,772 questions about drugs, indications, adverse events and mechanisms, scored for judged answer correctness and for citing the primary regulatory or peer-reviewed record behind the gold source. Evidence Index from 0 to 1 (0.4 judge correctness, 0.6 citation score, Llama-3.1-70B judge) for a model answering without tools; higher is better.
Metric: Evidence Index (0-1; 0.4 x judge-mediated answer correctness + 0.6 x authority-aware citation score, Llama-3.1-70B-Instruct judge, unparseable verdicts dropped; 3,772 items; model without tools). Source: arxiv.org. Saturation forecast: Around February 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 0.51 | #163 |
| 2 | GPT-5 Mini | 0.43 | #349 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=drugaudit · How It Works · Data refreshed daily, snapshot 2026-11-12.