GPT-6 Astra: from a chat response to completed work

OpenAI introduced GPT-6 Astra on 3 September. Its announcement highlights improvements in coding, research, computer use and more demanding multi-step tasks. The model is said to create documents, spreadsheets and presentations using the user's templates and adapt to additional requirements as the work progresses. [1]

For businesses, the main point of interest is a shift in expectations: not just “explain how”, but “prepare a result I can review and use”.

What could this mean in the office?

Imagine preparing a comparison of three supplier quotes. The desired result is not a long chat response, but a spreadsheet: comparable items, prices, deadlines and clearly marked information that is missing from the quotes. Or preparing materials for a meeting: selected documents become a brief summary and a presentation using the company's template.

These are examples of sensible tasks to try, not a report on our testing of Astra. This is precisely the kind of work where we would check whether the model transfers numbers correctly, preserves the meaning of the originals and flags missing information rather than filling it in on its own.

An introduction does not mean access for everyone

At the time of the announcement, OpenAI stated that it was rolling out access to a limited set of organizations and that Astra was not yet generally available. It announced broader availability over the following days. [1] This reflects the situation when the news was announced, not a guarantee of access on any particular plan today.

The company also announced additional safety oversight intended to detect cases where the agent may have misunderstood the user's instructions. [1] Such oversight is no reason to skip reviewing the finished work.

Synesis perspective

The most useful first step is not to hand the entire office over to AI. Choose one recurring task, define the input documents and the format of the result, and check how much correction is still needed. To begin with, have the system prepare a draft rather than automatically send quotes or change business records.

The true measure of progress is not the length of the response, but the time it takes to reach a usable and verified result. Astra is interesting because of the direction OpenAI emphasizes in its announcement; its practical value, however, needs to be judged through actual work.