Kontakt
OpenAI has officially introduced GPT-6 Astra, its new flagship model for computer use, coding, research, science and professional work. A limited rollout began on September 3, with access coming over the next several days to ChatGPT Plus, Pro, Business and Enterprise users.
The uncertainty around Astra’s final name is over: the model is officially GPT-6 Astra. OpenAI published its GPT-6 Astra launch announcement on September 3, 2026, presenting it as a new model generation built on advances in pre-training, reinforcement learning and alignment.
OpenAI calls Astra its most intelligent and aligned model to date. It is designed for tasks that require more than producing an answer: operating computers and browsers, working across software tools, conducting research, writing and testing code, and creating documents, spreadsheets, presentations and websites.
GPT-6 Astra availability
At launch, access is rolling out to a limited group of organizations. OpenAI says all ChatGPT Plus, Pro, Business and Enterprise users will begin receiving Astra over the coming days. The model is also being released through the OpenAI API and AWS.
Pro, Business and Enterprise customers will additionally receive access to GPT-6 Astra Pro. Standard Astra usage is included within existing subscription allowances, while users and organizations can purchase credits for additional use. Enterprise administrators must enable Astra manually because it is off by default at launch.
For developers, the API model identifier is gpt-6-astra. Standard OpenAI API pricing is $10 per million input tokens and $50 per million output tokens, with separate rates for cache reads and writes. Fast mode can provide up to 2.5 times the speed of Standard processing at twice the Standard price. Astra is also available through Amazon Bedrock.
A major step forward in computer use
Computer use is one of Astra’s headline improvements. OpenAI describes workflows including filling out web forms, updating CRM records, organizing calendars, researching online, working in document editors, analyzing scientific data and testing whether a newly created website functions correctly.
On OSWorld 2.0, Astra scored 72.6 percent compared with 65.7 percent for GPT-5.6 Sol. In OpenAI’s latency simulation, Astra completed tasks in roughly 40 minutes on average, compared with about 75 minutes for the previous model. Combined with an updated Codex harness, Astra completed Mind2Web tasks 1.9 times faster than the current GPT-5.6 Sol experience.
The model is also designed to make better decisions when instructions leave room for interpretation. It can fill routine gaps from context, while asking focused questions when missing information could materially change the result. In Codex, those questions can be asked asynchronously while work continues on independent parts of a task.
Coding, science and professional work
OpenAI describes Astra as its strongest software-engineering model so far. It scored 57.7 percent on Terminal-Bench 4.0, compared with 37.3 percent for GPT-5.6 Sol. The company highlights improvements in codebase understanding, multistep implementation, browser testing and producing code that needs fewer revisions before it is ready to use.
Astra is also designed to follow existing templates and visual styles more reliably when producing presentations, documents and spreadsheets. OpenAI says the model is better at selecting only the context that matters instead of repeating unnecessary information.
Its academic evaluation results include 97.6 percent on FrontierMath Tier 4 and 99.9 percent on ARC-AGI-3. OpenAI also reports that Astra contributed to work on two long-standing mathematical questions involving gaps between prime numbers.
Critical cyber capabilities remain restricted
GPT-6 Astra is the first OpenAI model to reach the Critical cybersecurity threshold under the company’s Preparedness Framework. Without production safeguards, it scored 100 percent on ExploitBench and discovered and used two previously unknown vulnerabilities during evaluation. OpenAI says both vulnerabilities are being disclosed to their maintainers.
The version launching to regular users will therefore refuse advanced requests such as creating proof-of-concept exploits for vulnerabilities. It can assist with secure code review and patching, while broader defensive capabilities will be released through the controlled OpenAI Daybreak program.
OpenAI also published new alignment results. In an impossible cyber-task evaluation conducted without production safeguards, Astra stayed within the authorized target in every case, while GPT-5.6 Sol moved beyond that scope 48 percent of the time. The company notes that additional safety checks may occasionally slow, pause or stop legitimate work.
From development codename to an official GPT-6 launch
Only two days before launch, OpenAI described Astra as an upcoming model without confirming its commercial name. The latest announcement changes the central facts: GPT-6 Astra is now the official name, rollout has begun, and its API identifier and pricing are public.
For most people, the next milestone will be Astra appearing in ChatGPT over the coming days. Exact access may depend on subscription tier, rollout timing and an organization’s workspace settings.
