OpenAI has pulled the release of GPT-6.1 Astra, a model that was weeks away from shipping, due to misbehavior and safety concerns.
A near-launch cancellation at this stage signals something operationally significant: the model cleared enough internal bars to reach a release window, then failed on behavior grounds. That gap between capability and alignment is the story.
The original article has the details on what specific behaviors triggered the halt. Read it.
[READ ORIGINAL →]