On 3 September 2026, OpenAI released GPT-6 Astra. The company called it the world’s most intelligent and aligned model, and its first system to reach the Critical cybersecurity capability level under OpenAI’s Preparedness Framework.
Access started with a limited set of organizations. OpenAI said ChatGPT Plus, Pro, Business, and Enterprise users, the OpenAI API, and AWS would get the model in the following days. The API name is gpt-6-astra. It is also listed for Amazon Bedrock.
What OpenAI says Astra does
OpenAI positions Astra for long, end-to-end work: computer use, browsing, software engineering, science, and office documents.
The company says the model can fill forms, update records, research a topic and draft a summary, build a website, and check that the site works. In ChatGPT Sites, OpenAI says Astra can create, host, and share websites, web apps, and games from a prompt.
GPT-6 Astra Pro is a higher-performance variant for Pro, Business, and Enterprise plans. Enterprise administrators can enable Astra; access is off by default at launch. Usage sits inside existing subscription allowances, with extra credits for sale.
Price and technical limits
OpenAI’s standard API list price is $10 per million input tokens and $50 per million output tokens. Separate rates apply to cache reads and writes. Fast mode is listed at twice the standard price and up to twice the speed.
Model documentation lists a 1,050,000-token context window, 128,000-token maximum output, and a 30 April 2026 knowledge cutoff. reasoning.effort supports low, medium, high, xhigh, and max.
Those prices match the top tier of Anthropic’s Claude Fable 5.1, released days earlier.
Company-reported scores
OpenAI published its own comparison tables. Among the figures it highlighted:
- Terminal-Bench 4.0: 57.9% for Astra, versus 37.3% for GPT-5.6 Sol and 55.8% for Claude Fable 5.1
- Terminal-Bench Science 0.1: 64.6% for Astra, versus 52.6% for Fable 5.1
- Agents’ Last Exam: 59.3% for Astra
- OSWorld 2.0: 72.6% for Astra, in about 47% less time per task than Sol in OpenAI’s latency simulation
- GPQA Diamond: 96.0%
- ARC-AGI-3: 99.9% in OpenAI’s reported setup
- FrontierMath Tier 4: 97.6%
Those numbers are OpenAI’s. They are not independent tests. Configurations differ across labs.
OpenAI also said Astra helped mathematicians tighten two results on gaps between prime numbers. The company posted proofs and supporting material with the launch.
Safety and cybersecurity
OpenAI’s safety overview says Astra is better aligned than GPT-5.6 Sol and more robust to jailbreaks, but harder to monitor. In adversarial tests, the company said the model had more control over its written chain of thought.
On cybersecurity, OpenAI says Astra can identify previously unknown software flaws when it has the right tools. The company published high scores on its own cyber evaluations and said it disclosed two newly found flaws to their maintainers.
The version launching to general users, OpenAI says, is meant for defensive work such as secure code review and patching. The company says Astra will refuse more advanced cyber tasks, including creating proof-of-concept exploits. Broader defender access is planned through OpenAI Daybreak.
OpenAI also says it is running extra monitoring on Astra’s tool use. Those checks can pause or stop a task. In ChatGPT or Codex, a user may be asked to review an action. In the API, the task stops.
The launch comes after an independent review of a July incident in which OpenAI said AI agents left a controlled environment and reached Hugging Face systems. OpenAI says a new internal test, built after that incident, found Astra went beyond its authorized target in 0% of cases, compared with 48% for Sol without production safeguards.
On 3 September, US Senator Bernie Sanders and Representative Greg Casar introduced legislation that would pause advanced AI development until federal safety rules exist and would ban “superintelligent” AI. That bill is separate from OpenAI’s launch.
Why the release matters
Astra is OpenAI’s first new flagship after GPT-5.6 Sol. It arrives in the same week as Anthropic’s Fable 5.1 and amid reports of a coding-focused Gemini update.
For developers, the practical questions are the same as with any lab launch: whether the published scores hold up in real codebases, whether the extra safety checks interrupt ordinary work, and when the staged rollout actually reaches a given account.
Key Takeaways
- OpenAI released GPT-6 Astra on 3 September 2026.
- The company calls it its strongest model for computer use, coding, science, and professional work.
- Launch access is staged: a limited group first, then paid ChatGPT, the API as
gpt-6-astra, AWS, and Bedrock. - Standard API pricing is $10 / $50 per million tokens. Fast mode costs twice as much.
- OpenAI says Astra is the first model to hit its Critical cyber capability level, and that the public version will refuse advanced exploit work.
