OpenAI releases MentalHealthBench, an expert-informed benchmark for AI responses in mental health conversations

What happened

- OpenAI News introduced MentalHealthBench, described as an expert-informed benchmark for evaluating helpful and safe AI responses across realistic mental health conversations (OpenAI News). The announcement states the benchmark's purpose but not its scores, methodology details, or which models were tested, so no performance conclusions can be drawn from this evidence. It matters because it gives the field a named evaluation target for a sensitive domain; whether the benchmark holds up under outside scrutiny is not established by the supplied evidence. [1]