OpenAI said on September 23 it released MentalHealthBench, an open benchmark for testing how AI systems respond to realistic mental health conversations.

The company said it co-created the benchmark with more than 80 licensed mental health experts from 22 countries.

The benchmark uses synthetic chats that cover non-acute conversations, high-acuity cases, and emergencies.

It includes adults, teens, caregivers, and clinicians across multiple languages and regions.

OpenAI said it scored model responses against expert criteria using an automated grader called GPT-5.6 Sol.

OpenAI also said a separate study involved 44 adults from 16 countries who had used AI for mental health or emotional support.

The company said participants valued tone and practical next steps more, while experts put more weight on gathering context and carefully interpreting ambiguous situations.

The release follows OpenAI separately strengthening responses in sensitive conversations and introducing Trusted Contact in ChatGPT.

It also comes amid wider scrutiny, with the US Food and Drug Administration issuing guidance on digital mental health medical devices and early lawsuits alleging that some AI chatbots encouraged minors’ suicides.

🔗 Source: OpenAI

Recent OpenAI developments