HELM MedHELM - MedHallu: leaderboard
HELM MedHELM run of MedHallu, detecting hallucinated content in medical question answering (exact match).
Metric: EM. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 92.6 | #291 |
| 2 | Gemini 2.0 Flash | 90.8 | #342 |
| 3 | GPT-5 | 90.5 | #98 |
| 4 | O3 Mini (2025-01-31) | 89.6 | #240 |
| 5 | Gemini 1.5 Pro (001) | 89.1 | #444 |
| 6 | Llama 3.3 70B Instruct | 88.2 | #534 |
| 7 | Claude 3.7 Sonnet (20250219) | 87.7 | #201 |
| 8 | GPT-5 Mini (2025-08-07) | 87.4 | #172 |
| 9 | O4 Mini (2025-04-16) | 86.8 | #179 |
| 10 | Gemini 2.5 Pro (Preview 05-06) | 86.4 | #147 |
| 11 | DeepSeek R1 | 84.7 | #251 |
| 12 | GPT-4o (2024-05-13) | 84.6 | #325 |
| 13 | GPT-4o Mini (2024-07-18) | 80.1 | #558 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-medhelm-medhallu · How It Works · Data refreshed daily, snapshot 2026-10-20.