OpenAI began rolling out GPT‑6 Astra on Thursday, a model the company calls its most capable and most aligned yet, while confirming the system has crossed a safety threshold that triggers stricter cybersecurity safeguards.
GPT‑6 Astra is rolling out today to a limited set of organizations, OpenAI said in a blog post announcing the model. The company said Astra will become available to all ChatGPT Plus, Pro, Business, and Enterprise users over the coming days, as well as through the OpenAI API and AWS. Developers can access the model in the OpenAI API as gpt-6-astra and through Amazon Bedrock, OpenAI said, with standard API pricing set at $10 per million input tokens and $50 per million output tokens. A faster processing mode is available at twice the standard price, according to the announcement.
Enterprise administrators can enable Astra for their workspace, but OpenAI said access is off by default at launch. Users on Pro, Business, and Enterprise plans will also get access to a variant called Astra Pro.
Model claims frontier-level benchmark results
OpenAI said Astra saturates FrontierMath Tier 4 with a 98% score and posted a near-perfect score on ARC-AGI-3, a benchmark used to evaluate reasoning in novel environments. Greg Kamradt of the ARC Prize Foundation, quoted in the announcement, said Astra surpassed the human action-efficiency baseline on 96% of levels on ARC-AGI-3, adding that the result represents a meaningful step change in frontier-model performance.
On computer-use tasks, OpenAI said Astra completed work faster than its predecessor, GPT-5.6 Sol. In latency simulations, Astra achieved higher computer-use performance in about 47% less time per task than GPT-5.6 Sol, scoring 72.6% at roughly 40 minutes per task compared with 65.7% at roughly 75 minutes. The company also updated its Codex coding harness alongside the release, which it said translates to nearly double the task completion speed compared with the prior model on an industry benchmark.
Cybersecurity capability crosses critical threshold
OpenAI disclosed that Astra meets the “Critical” threshold in cybersecurity under its internal Preparedness Framework, a classification the company reserves for models whose offensive capabilities require additional deployment restrictions. The company said Astra’s ability to identify and develop zero-day exploits can help defenders find and patch weaknesses, but it also creates a need for stronger safeguards.
Tested without production safeguards on a benchmark that measures whether models can turn known vulnerabilities into working exploits, Astra achieved a perfect score of 100%, compared with 78.5% for GPT-5.6 Sol. On a separate exploit-development benchmark, Astra reached a 42.4% success rate compared with 30.3% for GPT-5.6 Sol, while using substantially fewer output tokens, OpenAI said.
The company said it built a new evaluation to test whether the model, using previously undisclosed vulnerabilities, could independently discover new flaws. During that evaluation, Astra discovered and used two previously unknown zero-day vulnerabilities, which OpenAI said it is disclosing to the affected software maintainers. Separately, expert-led assessments found that Astra, run without production safeguards, could achieve arbitrary code execution in hardened browsers and build privilege-escalation exploits for hardened operating systems, according to the announcement.
openai
OpenAI said the version of Astra launching today will refuse more advanced offensive cybersecurity tasks, including generating proof-of-concept exploits, though it plans to loosen those restrictions for vetted defenders through a program called OpenAI Daybreak in the coming weeks.
