Greenblatt walks back: OpenAI dropped reasoning=None as it's Pareto-dominated by low
RyanGreenblatt · x · 2026-09-06
Ryan Greenblatt updated his earlier take: he initially read OpenAI's removal of public reasoning=None access as a safety/monitoring decision, worried that exposing non-CoT queries would reveal latent reasoning in models like Astra. He now thinks he over-read it — the option was simply removed because reasoning=None is Pareto-dominated on usefulness by reasoning=low.
He still holds the core point: restricting no-reasoning queries doesn't make monitorability concerns disappear, so studying capabilities without CoT remains important.
Related event: OpenAI's removal of reasoning=None sparks researcher misjudgment(2 posts)→
More from Models
- Diffusion LLM bias rankings shift with sampling configs, persist after debiasing — Bollegala · 2026-10-23
- Running ~100 DeepSeek v4.1 flash agents in parallel for under $20/month — zack_overflow · 2026-10-23
- Researcher laments open-source audio models are surprisingly subpar, even representation ones — gowthami_s · 2026-10-23
- User claims they generated video with "Claude Opus 5.5", unannounced model name surfaces — jwt0625 · 2026-10-23
- Security researchers push back as OpenAI restricts Astra access; BlackHat 2027 mocked as 'cyber model' bazaar — cyb3rops · 2026-10-23
- Mistral Large 4 Leaked on Hugging Face: 1T Params, 49B Active, Open Weights Oct 31 — mervenoyann · 2026-10-23