OpenAI launches GPT-6 Astra for longer workflows

GPT-6 Astra is meant to fill in forms, update records in a customer database and lay out presentations using a template that already exists. OpenAI released the model to a selected group on 3 September and describes it as the company's best so far at using a computer. It works in the browser, the spreadsheet and the desktop applications much as a person does.

The selling point is that the model takes the task all the way. OpenAI describes longer continuous workflows where Astra holds on to the original assignment when new instructions arrive, asks a question when the answer actually changes the outcome, and hands over a finished document instead of a description of how to produce one yourself.

Two other announcements sit inside the same launch. According to OpenAI the model has contributed to solving mathematical problems that had stayed open for a long time. In addition it is the first OpenAI model to reach the Critical level for cybersecurity in the company's "Preparedness Framework", which means that with the right tools and access it can find unknown vulnerabilities and build working attacks against well protected systems. For Enterprise and Edu accounts access is off by default.

The rollout starts with a limited set of organisations. Over the coming days the model is to reach Plus, Pro, Business and Enterprise as well as the API, where standard mode costs 10 dollars per million tokens in and 50 dollars out. So far the claim about finished documents instead of descriptions rests mainly on OpenAI's own measurements, made by the party selling the model.

If the claim holds, what is worth knowing moves. Operating the software becomes worth less when the model clicks its way through it. Judging whether what comes out holds up becomes worth more, and that judgement should not be left to the same model that did the work.