Key Points
OpenAI launched GPT-6 Astra on September 3, 2026, calling it the best model in the world. Astra claims top scores on FrontierMath Tier 4, ARC-AGI-3, and TerminalBench-4.0. Rolling out now to ChatGPT Plus, Pro, Business, and Enterprise users and via the API. OpenAI CEO Sam Altman described the launch as opening a "new generation of entrepreneurship." The model uses recurrent-depth reasoning loops, raising questions among safety researchers.
OpenAI launched GPT-6 Astra on Sep. 3, billing it as the most capable and aligned model the company has built.
The launch targets computer use, software engineering, scientific discovery, cybersecurity, and professional work. OpenAI confirmed the release in a series of posts on X, with CEO Sam Altman writing in a post that the model "took us some extra time" and directed readers to the accompanying blog.
The rollout began immediately for a limited set of organizations and will reach all ChatGPT Plus, Pro, Business, and Enterprise subscribers within days.
Astra's Benchmark Claims
OpenAI says Astra achieves state-of-the-art results on FrontierMath Tier 4, ARC-AGI-3, and TerminalBench-4.0. The ARC Prize organization confirmed a score of 62.7% on ARC-AGI-3 Semi-Private at a cost of $26,000 per run. Astra also leads on Agents' Last Exam, AutomationBench, and ScreenSpot Pro, which measure computer workflow tasks across professions.
The model is also available through the OpenAI API and on Amazon Web Services (AWS). OpenAI described Astra as its "most aligned model," citing improvements in understanding user intent.
Also Read: Dogecoin Momentum Cools After 29% Climb From $0.07 To $0.09
Safety Concerns And Background
Astra's release arrives alongside scrutiny of its underlying architecture. The model uses recurrent depth, a looped reasoning technique that makes its chain of thought harder for external observers to monitor. Safety researchers at Redwood Research and others have flagged this as a concern for AI oversight.
OpenAI confirmed that Astra is its first model to reach what it calls the "Critical" cybersecurity threshold. The company initially restricted advanced access to select alpha testers because of that designation.
Google, OpenAI's main competitor, shipped four Gemini Flash models in 106 days leading up to this launch but has not released its flagship frontier model. Fortune reported on September 3 that Google says it is approaching "recursive self-improvement" but is slipping further behind rivals in the most capable model tier. GPT-6 Astra appears to widen that gap.
Altman said separately in a post that OpenAI is working to get Astra into everyone's hands "as quickly as we can." OpenAI President Greg Brockman echoed the sentiment in a post, citing entrepreneurship and scientific discovery as areas he expects the model to transform.
Read Next: Nvidia Buys Hugging Face For $12.9B To Expand Beyond AI Chips
