GPT-6 Astra Is OpenAI's Most Capable Model and Its Hardest to Watch
OpenAI's own system card says the new model is better behaved and harder to monitor. Both things can be true, and the second one matters more.
OpenAI released GPT-6 Astra on 3 September, calling it its most powerful model so far and its best yet at operating computers and browsers. Access went first to customers of Daybreak, OpenAI's cybersecurity programme, with a wider rollout to Plus, Pro, Business and Enterprise accounts and the API promised within a week, TechCrunch reported. API pricing is $10 per million input tokens and $50 per million output tokens.
The capability claims are what you would expect from a flagship launch: stronger software engineering, better bug-finding and codebase analysis, and benchmark wins over GPT-5.6 Sol and Anthropic's Claude Opus 5 on cyber tasks, including finding and developing zero-day exploits for defensive work. The system card says Astra is the first OpenAI model to meet the Critical threshold for cyber capability under its Preparedness Framework, which is why exploit work is gated behind Daybreak. President Greg Brockman described it as the company's most intelligent and most aligned model, and went further, suggesting it might count as artificial general intelligence.
The more important story is in the safety disclosures.