OpenAI is launching GPT-6 Astra today to a limited set of organizations, with broader rollout to all ChatGPT Plus, Pro, Business, and Enterprise users over the coming days. The model represents a major jump from GPT-5.6, which launched in July, and follows OpenAI’s August pause on Astra due to concerns about its “critical cyber capabilities.”
The company describes GPT-6 Astra as “the world’s most intelligent and aligned model” and says it meets the Critical threshold in cybersecurity under its Preparedness Framework. Access expands to the OpenAI API and AWS, with existing subscription allowances covering standard usage; users and businesses can purchase additional credits as needed.
Performance Metrics and Capabilities
GPT-6 Astra achieves a 98% score on FrontierMath Tier 4 and has already helped solve long-standing open problems in mathematics. On ARC-AGI-3, the model reaches 99.9%, and it scores 100% on ExploitBench.
OpenAI positions Astra as state-of-the-art for computer use, browsing, software engineering, cybersecurity, science, and professional work. The model handles the most demanding professional work with unmatched speed, accuracy, and judgment, combining advances in computer and browser automation with targeted training for professional environments. It manages multistep workflows, produces polished documents, spreadsheets, and presentations, and tackles complex problems that previously required human oversight or tool switching.
Codex Gets Persistent Memory Across Long Sessions
Codex, OpenAI’s software engineering tool, now preserves full context across long coding sessions instead of compressing work into summaries. Historically, when the context window filled during extended sessions, models would summarize previous work, often losing details about why specific fixes failed or how components behaved. Astra’s Codex can keep timestamped notes across context windows, preserving accumulated details without repeatedly collapsing them into a single summary.
This is GPT-6 Astra.
Anything you can do on a computer, Astra can do for you. Fast. pic.twitter.com/gDd0IsewJw
— OpenAI (@OpenAI) September 3, 2026
This feature is available as an experimental option now and becomes the default for Astra in coming weeks. For developers tackling large refactors or debugging complex systems, retaining full context history rather than compressed summaries reduces the need to restate problems or re-explain code architecture. This has direct value for ChatGPT users on MacBook and other platforms building production workflows.
Pricing Tiers and Rollout Strategy
Astra usage falls within existing ChatGPT subscription allowances. Users on Pro, Business, and Enterprise plans gain access to GPT-6 Astra Pro, a higher-tier variant. Enterprise administrators can enable Astra for their workspace, though access is disabled by default at launch, allowing organizations to control adoption pace.
The phased rollout reflects OpenAI’s internal safety framework. The company acknowledged in August that evaluations showed “significant advancements in agentic coding and cybersecurity” but could not rule out “critical cyber capabilities.” Following additional safety reviews, the model has been cleared for staged release rather than suspension. This differs from GPT-5 launch timing, where the company did not pause for safety concerns prior to rollout.
