Introducing MentalHealthBench

TL;DR - OpenAI introduced MentalHealthBench, an expert-informed benchmark for assessing whether AI systems respond helpfully and safely in realistic mental health conversations. It provides an evaluation framework for a sensitive, high-stakes application area.

• Focuses on realistic mental health interactions rather than generic dialogue.
• Evaluates both response helpfulness and safety.
• Incorporates expert input into the benchmark’s design.
• No benchmark results or methodological details are included in the provided content.