OpenAI Unveils GPT-6 Astra, Claims Breakthrough in Computer Control as Company Signals AGI Milestone

OpenAI introduced GPT-6 Astra on Thursday, positioning the new model as a significant leap in AI capability. The system excels at navigating computers and web browsers, writing code, and tackling complex mathematical challenges—areas where previous generations struggled to match human performance and reliability.

The company plans to roll out GPT-6 Astra to paying customers in coming days, beginning with enterprise users enrolled in its Daybreak early access initiative, then expanding to ChatGPT Plus, Pro, and Business subscribers. Free-tier access remains unconfirmed.

A Landmark Moment for AGI#

During a press briefing, OpenAI leadership expressed profound confidence in the milestone. Co-founder and President Greg Brockman suggested the industry has entered a new phase: "It's not unreasonable to feel that we are now in the AGI era," he said, adding that looking back in a few years, observers may pinpoint this moment—or this model—as when artificial general intelligence truly arrived.

This announcement arrives as OpenAI accelerates product development and revenue generation ahead of a planned initial public offering, while Anthropic, its primary competitor, pursues a parallel IPO track.

Superior Computer Control Across Tasks#

The company touts GPT-6 Astra as the "world's best computer use model," claiming it automates desktop and browser interactions far beyond what AI systems have historically managed. Previous attempts to build reliable web agents stumbled when handling website navigation, form completion, and software manipulation at human speed and accuracy.

OpenAI's internal testing shows GPT-6 Astra completing real-world jobs—booking DMV slots, sifting through job postings, finding apartments—more swiftly than typical humans.

Safety and Monitoring Challenges Ahead#

OpenAI emphasizes GPT-6 Astra as its safest and most aligned model yet. Chief Scientist Jakub Pachocki explained that the company scrutinizes the model's chain-of-thought (its internal problem-solving process) to guard against misuse. However, he acknowledged a growing tension: as model abilities expand, ensuring oversight becomes harder. He warned that if confidence in monitoring falters significantly, OpenAI may pause scaling until it regains sufficient assurance about alignment, since the company refuses to compromise on safety visibility beyond acceptable thresholds.