OpenAI introduced MentalHealthBench on September 23, an open evaluation developed with more than 80 licensed mental-health experts from 22 countries. It uses synthetic conversations spanning everyday concerns, distress and emergencies, with expert-written criteria for helpful behavior and avoiding harm.
The scenarios include adults, teenagers, caregivers and clinicians. An automated grader assesses responses against expert-agreed rubrics. The benchmark supports research comparisons; it does not establish clinical effectiveness or guarantee safety in individual conversations. OpenAI notes that ChatGPT cannot replace professional care. Research announcement.