OpenAI releases MentalHealthBench, an open benchmark built with 80+ clinicians
coherence · x · 2026-09-24
OpenAI has released MentalHealthBench, an open benchmark measuring how frontier models perform in realistic mental health conversations, built with input from over 80 mental health clinicians. The methods are open so other researchers can run their own evaluations.
Quoting Boris MPower: once something is measured well, models tend to saturate that metric within a year — and he hopes the same happens for mental health support, given models already power billions of users.
More from Models
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12
- Gemini re-renders images as text when vision is removed, dropping only 6% WSR — zacharynado · 2026-11-12
- Bare-harness coding benchmark: Fable 5.2 best overall, Gemini 4 Carbon best value — Semantic_meaning · 2026-11-12
- Google's Physical-Engineering's Last Exam: all 14 frontier models score below 50% — dl_weekly · 2026-11-12
- Reddit's 2026 LLM hall of fame: Opus 5.5 still unbeaten after 9 rival releases — G0dZylla · 2026-11-12
- Commit leak reveals unreleased Gemini 4 variants: Argon, Hydrogen and Helix (unconfirmed) — BlackHC · 2026-11-11