OpenAI releases GPT-6 Astra, calls it the best computer-use model yet

OpenAI dropped GPT-6 Astra on Thursday, billing the flagship successor as "the world's best computer-use model." The company says it outperforms predecessors at browser and OS navigation, code generation, and complex math — areas where earlier models struggled to deliver consistent reliability. In internal tests, the model booked DMV appointments, searched job listings, and scanned rental apartments faster than the average human, according to the official blog post.

Rollout is phased and paywalled. Organizations in the Daybreak early-access program get it first, followed by ChatGPT Plus, Pro, Business, and Enterprise subscribers. OpenAI isn't saying whether or when free-tier users will see the model.

"It's not unlikely we're already in the AGI era," president and co-founder Greg Brockman told reporters. He believes that looking back in two years, the creation of artificial general intelligence — systems broadly smarter than humans — will be tied to this period and this model specifically.

The launch arrives at a pivotal moment. OpenAI is accelerating capability releases and revenue growth ahead of a planned IPO while trying not to compromise on safety. Its chief rival, Anthropic, is also preparing a public offering.

Chief scientist Jakub Pachocki called GPT-6 Astra the safest, most aligned model the company has shipped, crediting Chain-of-Thought monitoring — the mechanism that surfaces the model's internal reasoning. But he acknowledged that oversight gets harder as capabilities scale, and could become a bottleneck for future development. OpenAI will hold off on further scaling until it regains sufficient confidence in its ability to monitor those processes.