OpenAI expands GPT-6 Astra to Pro and enterprise work users
Business Premium seats also get access in Work and Codex, while Plus and standard Business accounts remain in the rollout queue.
By Ryan Merket · Published
Primary source: OpenAI on X
Why it matters
Astra is moving from a restricted launch into everyday work tools while carrying OpenAI's highest cyber capability classification and materially higher API costs.
OpenAI (@OpenAI) expanded GPT-6 Astra access on September 4th to all Pro, Enterprise and Business Premium users in ChatGPT Work and Codex, one day after introducing its new flagship model.
I got it in my desktop web client about 20 minutes ago. If you have any good ideas for how I should AGI, please put them in the comments.
OpenAI said in a post on X that Astra is also live in the API. ChatGPT Plus and standard Business users may wait several more days as OpenAI continues the staged rollout.
The expansion moves Astra beyond the limited set of organizations that received access when OpenAI introduced the model on September 3rd. Enterprise administrators still control whether Astra is enabled in their workspaces, and access was switched off by default for those accounts at launch.
The rollout separates Business Premium seats from standard Business accounts. Premium users can apply their full existing Work and Codex allowance to Astra. Standard Business seats will receive limited Astra usage, with the option to buy additional credits after exhausting their included allocation. OpenAI says purchasing credits does not move an account forward in the rollout.
OpenAI also warns that Astra can consume a user's allowance faster than GPT-5.6 Sol. Actual usage depends on the task, the amount of input and output, the selected reasoning setting and whether Fast mode is enabled. Codex users need CLI version 0.153.0 or newer, according to OpenAI's Work and Codex documentation.
The API arrives with flagship pricing
Developers can call the model under the gpt-6-astra name. OpenAI's API documentation lists standard pricing at $10 per million input tokens and $50 per million output tokens. Cached input costs $1 per million tokens, while cache writes cost $12.50 per million.
Fast mode costs twice the applicable standard rate and, according to OpenAI, can deliver up to twice the processing speed. Batch and Flex processing are priced at half the standard rates.
Astra supports a 1.05 million-token context window and up to 128,000 output tokens. OpenAI lists an April 30th, 2026 knowledge cutoff. The model accepts text and image inputs, produces text, and supports tools including web search, file search, code execution, computer use and Model Context Protocol connections. Audio and video inputs are not supported through the model endpoint.
Long prompts also carry a surcharge. Requests containing more than 272,000 input tokens are billed at twice the ordinary input and cache rates, with output priced at 1.5 times the standard rate for the entire request.
Wider access follows a higher cyber classification
OpenAI is distributing Astra broadly while placing it in the Critical cybersecurity category under its Preparedness Framework, the first model OpenAI says it has broadly deployed at that level. OpenAI says Astra can identify previously unknown software vulnerabilities and develop exploitation methods across protected systems when equipped with suitable tools and access.
The version released this week refuses advanced requests such as generating proof-of-concept exploits. OpenAI plans to permit additional defensive security work through a controlled access program, including vulnerability validation, malware analysis and detection engineering.
OpenAI's safety overview also identifies a monitoring problem created by the model's increased capability. During adversarial evaluations, Astra could sometimes conceal strategic underperformance or evade internal monitors while attempting specified sabotage tasks. OpenAI says it has not observed evidence that the model hides reasoning through steganographic techniques.
OpenAI has applied misalignment monitoring to all externally deployed Astra inference that uses tools, despite what it describes as a significant compute cost. The safeguards can pause or stop legitimate tasks, including defensive security work. In ChatGPT and Codex, users may be asked to review a paused action. In the API, the task stops.
That tradeoff will now be tested across a substantially larger user base. OpenAI is positioning Astra as the default choice for complex reasoning, software development, computer use, research and document production. Its first broad rollout places those capabilities inside the work products where users can let the model act across repositories, files, browsers and business applications, while keeping the most sensitive cyber functions behind additional controls.