OpenAI launches GPT-6 Astra, claiming state of the art in computer use
OpenAI announced GPT-6 Astra on 3 September, claiming a new state of the art in computer use, science and cybersecurity, with every figure coming from the company's own measurements.
OpenAI announced GPT-6 Astra on 3 September, presenting it on the company's own X account with a video and a single promise: "Anything you can do on a computer, Astra can do for you. Fast." The pinned launch post had 488,684 views at the time of measurement; the model's announcement page carries the details.
OpenAI says the rollout started the same day with a limited set of organisations, and that over the coming days Astra will become available to all ChatGPT Plus, Pro, Business and Enterprise users, through the OpenAI API and on AWS. The company also directs users to the ChatGPT desktop app. Separately, OpenAI ended its partnership with Cursor on 29 August following Cursor's acquisition by SpaceX; Cursor's direct access to OpenAI models ends on 12 November.
Benchmark results, all from OpenAI
OpenAI calls Astra the most intelligent and aligned model in the world and says it sets a new state of the art for computer use, browsing, software engineering, cybersecurity, science and professional work. The company reports leading results on Agents' Last Exam, AutomationBench and ScreenSpot Pro, benchmarks for computer workflow tasks across professions, and on FrontierMath Tier 4, ARC-AGI 3 and TerminalBench-4.0. For scientific discovery it cites Terminal-Bench Science 0.1 and HealthBench Pro. OpenAI also says the model better understands user intent. Every one of these figures is the company's own.
The Critical threshold in cybersecurity
Two days before the launch, OpenAI said Astra "represents a significant advance in cybersecurity capability, reaching the Critical threshold under our Preparedness Framework." That framework is the company's internal scale, not an outside standard. In the same post OpenAI said it was previewing how it evaluates the model and how its safeguards advance alongside capability. The claim came five days after OpenAI joined Anthropic, AWS, Google, Microsoft and Oracle in a 27 August call to strengthen cyber defence.
The launch video opens with archive footage of "Put That There", the 1979-80 MIT Media Lab demonstration of operating a computer by speech and pointing, used courtesy of MIT Media Laboratory, Chris Schmandt and Eric Hulteen.
Open note: GlobalFeed AI's publishing engine runs on Anthropic's Claude model, and Anthropic is a direct competitor of OpenAI.
