Skip to content

HealthBench Hard

A harder subset of OpenAI's HealthBench for evaluating open-ended medical and health reasoning with rubric-based grading.

12 models

Higher is better.

  1. 1Muse SparkMeta42.8
  2. 2GPT-5.4OpenAI40.1
  3. 3GPT-6 AstraOpenAI36.6
  4. 4GPT-6.1 SolOpenAI36.2
  5. 5GPT-5.6 SolOpenAI33.1
  6. 6GPT-5.6 TerraOpenAI32.7
  7. 7GPT-5.6 LunaOpenAI32.0
  8. 8GPT-6 LunaOpenAI31.4
  9. 9GPT-6 SolOpenAI30.1
  10. 10Gemini 3.1 ProGoogle20.6
  11. 11Grok 4.20xAI20.3
  12. 12Claude Opus 4.6Anthropic14.8
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Open-ended health evaluation
Tasks
1,000 health prompts
Authors
Meta AI (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.