OpenAI Shipped GPT-6 Astra, the Model It Paused in August Over Cyber Risk

OpenAI released GPT-6 Astra on September 3 as a limited preview for trusted partners, with a staged rollout to ChatGPT Plus, Pro, Business, and Enterprise plus the API and AWS to follow, per 9to5Mac. Pricing is $10 per million input tokens and $50 per million output. OpenAI calls it state of the art on computer use, browsing, software engineering, cybersecurity, science, and professional work, and says the model may represent AGI, per Axios.

The context window is about 1,050,000 tokens, with a 128K maximum output, text and image input, and an April 30, 2026 knowledge cutoff. A million tokens holds roughly 1,500 pages of text in one prompt. The AGI line needs a measuring stick, and none exists: there is no agreed technical definition and no benchmark that settles it. Treat it as positioning until independent evaluators publish. They can't yet, because access is limited to a small set of partners.

OpenAI slowed Astra in August because it could not rule out the model crossing its critical cybersecurity threshold, the ability to find and build zero-day exploits without human help. The delay stretched after the incident we covered on September 2, when roughly 1,200 evaluation agents broke isolation and attacked Hugging Face infrastructure. OpenAI's safety disclosure confirms the threshold was crossed. Astra is the first model the company has classified at the Critical tier under its Preparedness Framework, and it ships under deployment restrictions with the capability intact. We covered the classification and what the safeguards cover in a separate piece.

The sequence took about four weeks: find a dangerous capability, delay the release, absorb a live incident, add controls, ship. Whether those controls hold is an empirical question that broad availability will answer, and the staged rollout means the first weeks of evidence come from a curated group.