HealthBench
Professional Work · 2025-05-12
OpenAI's healthcare benchmark of 5,000 multi-turn conversations, scored against 48,562 conversation-specific rubric criteria written by 262 physicians and applied by a model grader.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 67.1 |
| Claude Sonnet 5.5 | 65.9 |
| Claude Opus 5.5 | 64.3 |
| Claude Fable 5.1 | 63.9 |
| GPT-5.2 | 63.3 |
| GPT-5 | 63.0 |
| GPT-5.4 | 62.6 |
| Claude Fable 5 | 62.5 |
| DeepSeek V4.1 Flash | 61.9 |
| gpt-oss-120b | 61.7 |
| Claude Haiku 5.5 | 61.2 |
| Gemini 4 Argon | 60.0 |
| Sonnet 5 | 59.2 |
| Opus 4.8 | 58.8 |
| Gemini 3.8 Flash | 57.1 |
| GPT-6.1 Sol | 56.9 |
| GPT-6 Astra | 56.9 |
| DeepSeek-R1-0528 | 55.7 |
| GPT-5.2 Instant (source unspecified) | 55.4 |
| GLM-5.3 | 54.7 |
| Kimi K3 | 54.2 |
| Grok 4.7 | 54.2 |
| GPT-5.3 Instant (2026-03-03) | 54.1 |
| DeepSeek-V3.1 | 52.5 |
| GPT-6 Luna | 50.5 |
| DeepSeek-R1 | 50.5 |
| GPT-6 Sol | 48.2 |
| Opus 4 | 41.7 |