GPT-6 Astra ships after the red line
OpenAI’s next frontier model is rolling out — Daybreak first, then ChatGPT paid tiers and the API. Computer use and coding are the pitch. “AGI era” is the marketing. The August cyber pause is the context.
By Drew Wall,
On 3 September 2026 OpenAI started shipping GPT-6 Astra — the same frontier model it paused in August after saying it could not rule out Critical cybersecurity capability under its own Preparedness Framework. The product story is computer use, coding, and multistep agents. The press story is Greg Brockman calling this the start of an "AGI era." The operational story is a staged rollout: not everyone gets it today.
What is shipping
Access begins with enterprise cybersecurity customers on OpenAI's Daybreak program, then expands over the coming days to ChatGPT Plus, Pro, Business, and Enterprise, plus the OpenAI API and AWS. OpenAI is pitching Astra as a generational leap on software engineering, professional documents, and agents that drive real computer interfaces — less "chat about the task," more "do the task." Astra Pro is the higher tier for Pro, Business, and Enterprise. There is still no standalone Astra website to list; it lands inside ChatGPT and the API stack you already know.
The red line did not disappear
OpenAI now treats Astra as meeting its Critical cybersecurity threshold — the top rung that forced the August slowdown — and says trusted defenders get less-restrictive access for vulnerability work, malware analysis, and detection engineering. Safety messaging is heavy on purpose: after the summer Hugging Face intrusion, the company needs Astra to look powerful and supervised. Alignment leads talk about monitoring, rapid escalation, and the uncomfortable fact that smarter models are harder to read. Shipping after a public Critical pause is the point of the framework — if it means anything.
What to ignore, what to watch
Ignore the AGI slogan as a definition. Watch three concrete things: whether your ChatGPT or API workspace actually gets Astra this week; whether computer-use agents hold up outside demos; and whether Critical-tier access stays gated to defenders or quietly widens. For the stack around agents and models, see Agentic AI and Language Models.
The point
GPT-6 Astra is now a real product. It is rolling out in stages, carries a Critical cybersecurity rating, goes to Daybreak customers first, and is aimed at the enterprise computer-use work Anthropic has been winning. The question that matters is whether the safeguards that justified pausing it in August still hold once millions of paying users can use it.