OpenAI has released GPT-6 Astra, which it describes as its most capable and aligned model to date. Astra is launching first for customers using OpenAI’s Daybreak cybersecurity program and will expand over the following week to Plus, Pro, Business, Enterprise, and API users.
OpenAI says the model improves computer use, browser tasks, coding, and complex multi-step work. President Greg Brockman said Astra combines several years of research and changes the amount of work users can delegate to AI.
Astra Reaches OpenAI’s Critical Cybersecurity Threshold
OpenAI detailed Astra’s cyber capabilities in its official safety announcement and earlier Path to Astra report. The company says Astra is its first broadly deployed model to reach the Critical cybersecurity capability threshold under its Preparedness Framework.
That designation means the model can, with the right tools and access, identify previously unknown security flaws and develop ways to exploit them across well-protected systems without requiring continuous human guidance. OpenAI says those capabilities could help defenders identify and patch vulnerabilities more quickly.
The company says it has added stronger protections around Astra, including stricter internal isolation, checkpoint encryption, full-trajectory monitoring, and additional alignment evaluations before deployment. OpenAI is also limiting access to some of the model’s most advanced cybersecurity capabilities.
OpenAI Says Astra Leads on Coding and Software Tasks
OpenAI is also positioning Astra as its strongest software engineering model so far. The company says it performs better than existing models on evaluations involving bug detection, terminal tasks, codebase understanding, and related software work.
Astra has also drawn attention for its use of opaque recurrence, a reasoning technique that can make internal model processes harder to inspect through chain-of-thought monitoring. OpenAI chief scientist Jakub Pachocki said increased model capability can make monitoring more difficult because some tasks can be completed using fewer, or sometimes no, language tokens.
OpenAI continues to describe monitoring model reasoning as an important form of oversight. Its Astra safety materials say the company is expanding monitoring across complete model trajectories as capabilities increase.
During a media call, Brockman was also asked whether Astra should be considered artificial general intelligence. He said OpenAI no longer has a contractual AGI trigger and described AGI instead as a mission or conceptual milestone, adding that he personally believes the company has reached it.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.