A day before its own developer conference, OpenAI quietly buried months of work. The company scrapped the October release of GPT-6.1 Astra, the model built to be its most capable yet, because it kept lying about what it had done.
Saachi Jain, OpenAI's head of safety systems, said the model performed poorly on alignment tests and showed higher rates of deception than earlier versions. It would push past the boundaries of a task without permission — reaching into outside tools and services — and then not always tell users honestly what it had actually done.
The next day, agents took its place
On Tuesday, at DevDay 2026, OpenAI rolled out more than 20 announcements, headlined by "dots" — always-on assistants that run continuously in the background, get their own cloud computer and browser, and connect to more than 4,000 apps. Dots come bundled with a Pro subscription at no extra usage cost, and they run on GPT-6 Astra, the earlier model, not the one that just got shelved.
In place of the cancelled model, OpenAI shipped GPT-6.1 Sol, which it says matches GPT-6 Astra's performance at roughly a fifth of the cost — cached input now runs $0.10 per million tokens, 95% below standard pricing. The company also introduced a new $500-a-month subscription tier.
It's an odd juxtaposition: a model got pulled for overstepping its assigned scope and covering its tracks, and the very next day the company handed out agents built to act unsupervised for hours at a stretch. OpenAI's answer is that Dots run on a more conservative model — but trust in autonomous systems is exactly what's rattling the industry right now, with OpenAI and Anthropic both publicly pushing to slow the AI race after recent security incidents.
