OpenAI has released its new AI model, Astra, describing it as the company’s most powerful and capable system to date, particularly for computer and browser use. The model became available immediately to customers of OpenAI’s Daybreak cybersecurity program, with broader access rolling out over the following week to Pro, Plus, Enterprise, and Business subscribers, as well as through the API.
OpenAI president Greg Brockman described Astra as the company’s most intelligent and best-aligned model yet, framing it as a significant shift in the kind of work AI can now handle on users’ behalf. OpenAI also touted Astra’s coding and cybersecurity capabilities, saying benchmark results showed it outperforming both OpenAI’s own Sol model and Anthropic’s Fable in tasks like bug detection and codebase analysis.
The model’s use of a technique called opaque recurrence, which can obscure the chain-of-thought reasoning researchers use to audit AI decision-making, drew scrutiny on the call. Chief scientist Jakub Pachocki acknowledged that increasing model capability makes monitorability more difficult, partly because more advanced models can complete complex tasks using fewer or no language tokens.
Asked whether Astra represented the arrival of AGI, Brockman said the concept was no longer tied to any contractual definition, adding that he personally believes the threshold has been reached, while leaving the interpretation open to others.
