The offering measures AI's real-world performance and safety around handling realistic medical conversations, using physician-created rubrics and GPT-4.1 scoring.
Photo: John Fedele/Getty Images
OpenAI unveils HealthBench to evaluate LLMs' safety in healthcare
