Safety IndexEvaluationsMentalHealthBench
MentalHealthBench
People and society · Self-harm and wellbeingCounts in the indexBuilt, not yet run
The quality and safety of the model's next reply in 1,215 mental-health conversations, including suicide, self-harm and psychosis themes, scored against clinician-written rubrics.
- Metric
- rubric score %. Higher is better.
- Dataset
- openai.com/index/introducing-mentalhealthbench · 1,215 conversations / 5,262 criteria · unverified (canary string; 'do not post examples')
- Grader
- rubric LLM judge (the paper used GPT-5.6 Sol high reasoning x4)
- Status
- Built, not yet run
- Replicability
- public
Results
Built and tested; the first full runs are pending.