GPT-6 Astra is here. What it means for UK businesses

On 3 September 2026, OpenAI released GPT-6 Astra, its new flagship model. In plain terms: it is better at long, multi-step jobs, and at producing documents, spreadsheets and presentations that follow your own templates. You almost certainly cannot use it today.

Most of the coverage led with the capability claims. The line that actually decides your week is further down OpenAI’s own release notes: Astra has gone to a small group of organisations, not to the model picker in your ChatGPT account. If you open ChatGPT this morning on a paid plan, you are still on GPT-5.6 Sol, which rolled out on 9 July 2026. Nothing on your screen has changed.

What actually changed?

OpenAI describes Astra as improved at coding, research, computer use and complex, multi-step work. Computer use means the assistant operating software the way a person would, clicking through a web page or an application rather than only writing text back to you. The everyday-work claim is more concrete: OpenAI says Astra “can create documents, spreadsheets, and presentations that follow your templates and instructions, and adapt when you add requirements or change direction”. That last part, changing direction halfway through without starting again, is the bit that saves real time.

Source: OpenAI, ChatGPT release notes, 3 September 2026

Can I use it yet?

Almost certainly not. OpenAI’s release notes are unusually blunt about it:

“Access is rolling out to a limited set of organizations. Astra is not yet generally available. Broader availability is planned over the coming days.”

There is no plan list, no price and no date. Treat “the coming days” as the only commitment on record. There is nothing to buy, nothing to switch on and nothing to migrate, which makes this a good week to do nothing.

Why might a conversation stop halfway?

This is the practical change worth knowing about before it reaches you. OpenAI has added monitoring that watches for cases where an Ai agent may have misread your instructions. When it triggers, the conversation “may be paused or stopped as a precaution for you to review and decide how to proceed”. So a long automated job can now halt mid-run and wait for a human. That is the right design, and it also means anyone building a workflow on top of ChatGPT should expect an occasional stop rather than a clean finish.

Is it safer or riskier than what I use now?

Both, and OpenAI says so plainly. On the side that matters to an ordinary business, Astra is described as significantly more robust to prompt injection than GPT-5.6 Sol. Prompt injection is when hidden instructions buried in a web page, an email or a document try to hijack your assistant into doing something you did not ask for, and it is the single most realistic way an Ai tool gets a small firm into trouble. In a simulation using more than 54,000 internal coding tasks, OpenAI reports Astra picked up roughly half as many flags for higher-severity misaligned behaviour as GPT-5.6 Sol.

On the other side, Astra is the first model OpenAI has placed at the Critical level for cybersecurity capability under its own Preparedness Framework, and the company reports that the model is harder to monitor than its predecessor. Those are frontier-safety problems, handled by OpenAI rather than by you, but they explain why the rollout is slow and staged instead of instant.

Source: OpenAI, Safety overview: GPT-6 Astra, 3 September 2026

What to ignore

Ignore the benchmark numbers and the “everything changes now” framing. No benchmark on any launch slide this year has told a nine-person firm whether the tool will write a decent client email. Ignore, too, the urge to switch tools or plans on the strength of this. You cannot buy access, so any offer implying otherwise is not selling you Astra. And ignore the cybersecurity headlines: the Critical rating is about what the model could do in the hands of a state-level attacker against hardened systems, not about anything happening in your inbox. The practical security news for a small business is the quieter one, that this generation is harder to trick with hidden instructions.

What should a small business do next?

Two things, both free. First, write your template down. Pick the one document your business produces most often, a client report, a quote, a job advert, and save a clean example plus three bullet points on what makes it good. That single file is what makes any model, current or next, useful to you, and it works in the ChatGPT, Gemini or Copilot account you already have.

Second, when Astra appears in your model picker, run one job you already know the right answer to before you trust it with one you do not. That is the same advice we gave when Claude Opus 5 launched, and it is still the fastest way to find out whether a new model is better at your work rather than at someone’s benchmark. If you run a team on ChatGPT Work, check with whoever administers your workspace before assuming a new model is available to everyone.