OpenAI announced the release of its GPT-6 Astra model on Thursday, marking the start of a phased rollout for enterprises in its Trusted Access Program.

The company described Astra as its most capable model to date, with strong performance in computer use, browsing, software engineering, cybersecurity, science, and professional tasks. It achieved a 97.6 percent score on FrontierMath Tier 4, a 99.9 percent score on ARC-AGI-3, and a perfect 100 percent on ExploitBench.

Compared to the prior GPT-5.6 Sol model, Astra completed computer-use tasks in about 47 percent less time while scoring higher on benchmarks such as OSWorld 2.0. The model also showed improved alignment, with zero instances of exceeding authorized task boundaries in testing where the prior model did so nearly half the time.

Pricing starts at $10 per million input tokens and $50 per million output tokens for standard use. Access expands over coming days to ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as the OpenAI API and AWS.

OpenAI conducted its largest training run yet for the model, using more than 100,000 GPUs. The White House approved the release following review under the Trump administration’s voluntary framework.

Limited cybersecurity capabilities will initially go to testers through the Daybreak program, with broader defensive access planned later. Microsoft will offer Astra through its Foundry platform for enterprise customers.

The rollout follows OpenAI’s earlier assessment that Astra meets critical cybersecurity thresholds under its preparedness framework, prompting safeguards on advanced capabilities.