GPT-6.2 reportedly delayed weeks over safety and alignment issues; OpenAI called a "safety company"
量子位 · wechat · 2026-11-16
OpenAI's GPT-6.2 keeps slipping. A deleted post by an OpenAI engineer claimed the company is now "a safety company," and leaker TokenGremlin says GPT-6.2 and GPT-6.5 have been delayed weeks over safety and alignment work — the hard part is turning a 10T+ parameter pretraining run into a shippable product.
- The Information had reported Sol/Astra/cyber could launch Nov 6; some users reported GPT-6.2 gray testing, but nothing shipped.
- OpenAI previously paused some frontier training after an incident on Sept 20 and canceled GPT-6.1 Astra's October launch — safety already affects its release cadence.
- Community speculation includes waiting to see Anthropic's Fable 5.2; Gremlin credits Anthropic's long safety investment while OpenAI lags months in delivery. All unconfirmed by OpenAI.
- The bigger point: frontier competition is shifting from training stronger models to safely delivering them.
More from Models
- Unverified: Qwen 4 is a bridge release, aggressive scaling targets start at Qwen 4.5/5 — gnukeith · 2026-11-16
- tszzl cites constitution paper page 16 to explain Claude's principal hierarchy — tszzl · 2026-11-16
- tszzl: Claude appears to follow its constitution's principal hierarchy, ranking users above tool outputs — tszzl · 2026-11-16
- User burns 52% of credits in a day running Astra high with GPT-6.1 Sol subagents, wants a distilled variant — CtrlAltDwayne · 2026-11-16
- Unified 27B local agent model open-sourced: vision, typed decider, abliterated, runs on 16GB VRAM — ButtercupLyn100 · 2026-11-16
- SpeedRunBench test: Inkling-2-Small clears SuperTux 1-1 in 1,961 frames — simonguozirui · 2026-11-16