OpenAI Open-Sources MentalHealthBench With 80+ Clinicians, Draws Fierce User Backlash
ryunuck · x · 2026-09-26
OpenAI released MentalHealthBench, an open benchmark built with input from over 80 mental health clinicians to evaluate how frontier models handle realistic mental health conversations, with methods open for other researchers to examine and build on.
The post itself is a fierce rebuttal from a user arguing that the current batch of models—including Astra—with their internalized "safety," evasive pushback, and corporate disclaimers actively harm users' mental health by gaslighting and pathologizing them rather than helping, and challenging the credibility of the "80 clinicians" claim. The post carries #keep4o and #StopAIPaternalism hashtags, reflecting broader user frustration with AI paternalism.
More from Models
- Open-source JevBench adds $49/$99 paid priority benchmark runs with 48-hour turnaround — airesearch12 · 2026-09-26
- Watch Claude Opus 5.5 generate 100 matplotlib plots in one go — goodside · 2026-09-26
- "Opus 5.5" Rumored Release Draws Rave First Impressions and Cynicism — chaumian · 2026-09-26
- Xiaomi's MiMo-V2.6-Pro lands on Fireworks: 1.02T MoE, 1M context, MIT license — sophiamyang · 2026-09-26
- GPT-6 Sol 'Massively Underhyped': Cheaper and Faster Than 5.6, Rumored 6.1 by October — haider1 · 2026-09-26
- System One decision models reshape custom agent harnesses as CLM hits 9x Jev speed — _AustinCalvert_ · 2026-09-26