HELM Lite - MedQA: leaderboard
HELM Lite v1.13.0 MedQA run: English four-option USMLE-style medical licensing exam questions (503 evaluated), 5-shot, exact match on the chosen option. Percent correct, 0-100, higher is better; guessing scores 25.
Metric: Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 91 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 86.28 | #343 |
| 2 | Claude 3.5 Sonnet (20241022) | 85.88 | #302 |
| 3 | GPT-4o (2024-05-13) | 85.69 | #337 |
| 4 | Claude 3.5 Sonnet (20240620) | 82.5 | #308 |
| 5 | GPT-4 Turbo (Preview) | 81.71 | #416 |
| 6 | GPT-4 (0613) | 81.51 | #493 |
| 7 | Nova Pro | 81.11 | #594 |
| 8 | DeepSeek V3 | 80.91 | #329 |
| 9 | Llama 3.1 405B Instruct | 80.52 | #470 |
| 10 | GPT-4 Turbo | 78.33 | #417 |
| 11 | Llama 3 70B | 77.73 | #632 |
| 12 | Mistral Large 2 (Jul) | 77.53 | #490 |
| 13 | Claude 3 Opus (20240229) | 77.53 | #428 |
| 14 | Gemini 1.5 Pro (002) | 77.14 | #368 |
| 15 | Llama 3.1 70B Instruct | 76.94 | #567 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-lite-medqa · How It Works · Data refreshed daily, snapshot 2026-11-05.