OpenAI Launches GPT-6 Astra for Paid ChatGPT Users

OpenAI has introduced GPT-6 Astra, a new model built to handle coding, computer use and longer sequences of digital tasks. The rollout begins with enterprise customers in the company’s Daybreak early-access program before expanding to ChatGPT’s paid plans over the coming days.

Who gets GPT-6 Astra first

OpenAI plans to make Astra available to Plus, Pro, Business and Enterprise subscribers, while access through the OpenAI API and Amazon Web Services is also planned. The company has not announced whether free ChatGPT users will receive the model, leaving the initial launch focused on paying customers and developers.

The company is emphasizing what the model can do inside ordinary software. OpenAI says Astra can navigate websites, fill out forms and complete multi-step assignments, including building working sites and producing documents, spreadsheets and presentations. In demonstrations described to reporters, the model handled tasks such as searching for jobs, looking for apartments and booking appointments.

Those examples are company claims rather than proof that the system will perform every real-world task reliably. Computer-use agents must interpret changing interfaces, permission prompts and incomplete instructions, all of which can turn a polished demo into a more complicated experience for users.

Big capability claims meet a safety test

OpenAI has also designated Astra as its first model to meet what it calls a critical cybersecurity capability threshold. That label raises the stakes for access controls and monitoring because stronger defensive tools can also create new risks when used without authorization.

Company president Greg Brockman argued during a press briefing that the launch may be remembered as part of the arrival of artificial general intelligence. That is OpenAI’s interpretation, not an independently settled milestone. The phrase remains debated across the technology industry, and broad claims about human-level intelligence cannot be resolved by a product announcement alone.

Safety questions are therefore central to the release. OpenAI says Astra is its most aligned model yet, while chief scientist Jakub Pachocki acknowledged that monitoring becomes harder as systems grow more capable. He said the company would pause further scaling if it could not maintain enough confidence in that oversight.

For subscribers, the practical test will be simpler: whether Astra can complete useful work accurately while keeping people in control. Its staged release gives OpenAI a chance to watch enterprise use first, but the wider paid rollout will expose the model to a much broader mix of tasks, websites and expectations.