OpenAI Open-Sources MentalHealthBench Built With 80+ Clinicians

The notable move here is governance, not capability: a frontier lab is handing external researchers the evaluation methodology itself, not just a scoreboard, which lets third parties audit how mental health performance is defined and where models fail. That matters most for clinicians and researchers, who currently lack shared standards for judging these systems on everyday support rather than only crisis response. The open question is whether other labs adopt or contest the benchmark, and whether clinician-designed evaluations become the default reference point for this category.