OpenAI’s new flagship model can operate computers, write code and browse the web — but access is being introduced in stages, with safety controls built in from the start.
OpenAI has announced GPT-6 Astra, describing it in official material as “the world’s most intelligent and aligned model.” The company says the model is designed to carry out multi-step tasks on a computer — including browsing the web, writing and executing code, working with office software, and assisting with professional and scientific work. Access is not being opened to everyone at once.
The rollout is being introduced in phases. According to OpenAI’s official announcement and developer documentation, Astra is going first to a limited set of organisations, before being made available to ChatGPT Plus, Pro, Business and Enterprise subscribers, and through the OpenAI API, Microsoft Azure and AWS Bedrock.
What GPT-6 Astra Can Do
According to OpenAI’s product page and API documentation, Astra carries a context window of 1.05 million tokens — meaning it can process and reason over very large amounts of text, code or data in a single session. The model supports up to 128,000 output tokens and has a knowledge cutoff of 30 April 2026.
OpenAI says the model is built for computer use, software engineering, cybersecurity, science and professional work. Its release notes confirm it can create documents, spreadsheets and presentations. The company’s own promotional post stated that “anything you can do on a computer, Astra can do for you” — though that claim is marketing language from the announcement post and isn’t independently verifiable as a precise technical description of the model’s capabilities.
Pricing through the API is set at the equivalent of around £8 per million input tokens and around £40 per million output tokens, converted from OpenAI’s listed rates of US$10 and US$50 respectively.
The Safety Question
OpenAI has been unusually direct about the risks attached to this model.
According to its safety overview and Preparedness Framework documentation, GPT-6 Astra has reached the Critical cybersecurity capability level — the highest tier in OpenAI’s internal assessment system. That classification indicates the model is capable enough in security-related tasks that OpenAI considers it to warrant restricted access and active monitoring.
OpenAI’s Help Centre release notes say extra safety monitoring is in place, and that conversations may be paused or stopped if a potential issue is detected. The phased rollout, rather than a broad public launch, appears to reflect that caution directly.
Sam Altman, OpenAI’s chief executive, has previously described the company’s approach to powerful models as requiring safety work to keep pace with capability — though no specific Altman quote on GPT-6 Astra was available in the official material at the time of publication.
Agentic AI: The Bigger Picture
GPT-6 Astra sits within a broader industry push toward what developers call “agentic” AI — systems that don’t just answer questions but carry out sequences of actions on a user’s behalf. Rather than generating a response and stopping, an agentic model can open a browser, write and run code, fill in a form, or move between applications to complete a task.
OpenAI’s developer documentation frames Astra as built for exactly this kind of work. The model is designed to operate across code environments, web browsers and office software — essentially acting as an automated operator of the computer itself, rather than a conversational assistant sitting alongside one.
That’s a meaningful shift from earlier GPT models. But it also raises practical questions about oversight: if a model can take actions on a computer autonomously, what happens when it makes an error, or when a task involves sensitive data? OpenAI’s safety monitoring is intended to address that, though the documentation doesn’t detail every scenario in which it would intervene.
Who Gets Access, and When
The phased approach means most users won’t have immediate access. According to OpenAI’s announcement, organisations in the initial cohort will get the model first; broader availability for ChatGPT Plus, Pro, Business and Enterprise users, and through cloud platforms, follows after that.
For businesses and developers using Microsoft Azure or AWS Bedrock, access to Astra will depend on when OpenAI enables the model within those services. No specific dates for each phase were confirmed in the available material at the time of publication.
No official response from the UK government, the Information Commissioner’s Office, or any UK regulatory body was identified in the available material.
What This Means for Kent Residents
Kent residents and small businesses subscribed to ChatGPT Plus, Pro, Business or Enterprise plans should gain access to GPT-6 Astra once the rollout reaches those tiers, though the timeline hasn’t been confirmed. Local organisations already using Microsoft Azure or AWS Bedrock — whether in the private sector or public services — could also deploy the model once it becomes available through those platforms. No Kent council service or NHS Kent and Medway ICB scheme was identified in OpenAI’s material as part of any early access arrangement, so most public-sector use in the county will depend on broader UK rollout decisions made at a national level.
Source: @OpenAI
OpenAI Launches GPT-6 Astra with Computer-Use Capabilities and Phased Rollout Quiz
5 questions