Artificial Analysis launches HealthBench and HealthBench Professional leaderboards
Original titleHealthBench and HealthBench Professional: Evaluating models in realistic health conversations
Artificial Analysis launched leaderboards for HealthBench and HealthBench Professional, OpenAI benchmarks that grade model replies to health conversations against physician-created rubrics.
HealthBench Professional leads with GPT-6 Astra at 66.0% on the length-adjusted score, while Claude Sonnet 5.5 tops HealthBench at 62.1%.
The length adjustment lowers scores for replies over 2,000 characters, which pushes Claude Opus 5.5 down from third-highest raw score to sixth.
The report explains how each benchmark scores replies and how length adjustment changes the rankings, which helps readers interpret the leaderboards correctly.
Source: Artificial Analysis Articles · artificialanalysis.aiPublished · added here