OpenAI's GPT-6 Astra is rolling out in stages

OpenAI has started releasing its most powerful model so far. Cybersecurity partners get it first, with business and consumer plans following.

OpenAI began rolling out GPT-6 Astra on 3 September. The company calls it its most powerful system so far, and says it is state-of-the-art on computer use, browser use, software engineering, cybersecurity, science and professional work. Access is staged rather than switched on for everyone at once.

Companies in OpenAI's Daybreak cybersecurity program get it first. Broader enterprise and consumer access is due in the coming days. On ExploitGym, a cybersecurity test, Astra scored higher than its predecessor GPT-5.6 Sol while using fewer output tokens, so it did more with less generated text.

The claims around it are unusually big. OpenAI president Greg Brockman described the model as a jump in capability and said it qualifies as artificial general intelligence. That language matters less to a midsize business than the practical part: if Astra is genuinely better at operating a browser and a computer, the tasks you can hand to an AI agent get longer and less supervised.

The same release triggered OpenAI's own internal safety protections. The company added monitoring designed to rapidly detect and contain potentially misaligned actions, and Sam Altman confirmed the system went through voluntary White House vetting. Chief scientist Jakub Pachocki put the limit plainly: "As these models become more capable, understanding exactly what they can do gets harder." If your team is on a Business or Enterprise plan, the model underneath your tools may change without you choosing it, so it is worth re-testing any workflow you rely on before you assume it still behaves the same way.

Staged rollouts are useful. You get a few days to watch what other people hit before it lands in your account.