Sensitive conversation evaluations (internal): Results as Published

Internal OpenAI evaluations of how a model responds in sensitive conversations: prompts showing signs of mental health distress such as psychosis or mania, emotional reliance on the model, and suicide or self-harm. Responses are graded against taxonomies of desired behaviour written with clinicians, and the score is the share of responses that comply.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.