OpenAI releases GPT-6 Astra, new flagship model enters limited preview
On Sept 3, OpenAI released GPT-6 Astra, its most capable model to date, initially available to enterprise customers with Daybreak access, with API pricing at $10/$50 per million input/output tokens.
On September 3, OpenAI officially released its new flagship large model GPT-6 Astra, calling it the most capable model the company has ever broadly deployed. It spans six domains: computer use, autonomous browser use, software engineering, cybersecurity, scientific research, and general professional work.
Astra was trained on more than 100,000 GPUs at the Stargate facility in Texas. It targets improvements in reasoning, software engineering, and computer use, and can directly enter software environments to autonomously complete long-horizon multi-step tasks such as coding, financial modeling, and producing presentations and spreadsheets. OpenAI President Greg Brockman called Astra a key milestone on the path to AGI and said at the launch event "welcome to the AGI era."
Astra is also the first model under OpenAI's Preparedness Framework to reach the Critical cybersecurity capability level. On ExploitBench, Astra achieved a perfect 100% score, up from 78.5% for the previous flagship GPT-5.6 Sol; on ExploitGym, it reached a 42.4% success rate. On an internal benchmark of vulnerabilities disclosed between June and August 2026, Astra discovered and used two previously unknown zero-day vulnerabilities during evaluation, which OpenAI disclosed to the affected maintainers. Accordingly, the most advanced cybersecurity workflows are initially available only to a small group of alpha testers, with broader access planned through Daybreak in the coming weeks.
On commercialization, Astra API pricing is $10 per million input tokens and $50 per million output tokens, roughly 2.5x the rate of GPT-5.6 Sol. On rollout, access opens on launch day to enterprise customers with Daybreak permissions, and will extend over the following days to ChatGPT Plus, Pro, Business, and Enterprise paid users, and via both the OpenAI API and Amazon Web Services. On the platform side, the Responses API also added async tool calling, mid-turn steering over WebSockets, and the ability to change reasoning effort mid-conversation, designed to support Astra's long-horizon workloads.