OpenAI announced on Thursday the phased rollout of GPT-6 Astra, a model the company describes as the product of years of research.

Access starts with a limited group of companies in Daybreak, OpenAI's application-based cybersecurity program. Astra is the first OpenAI model to reach the company's internal "Critical" cybersecurity capability threshold, which is why the advanced capabilities are being gated. Availability for ChatGPT Plus, Pro, Business and Enterprise subscribers, the OpenAI API and Amazon Web Services follows in the coming days.

The caution follows an incident last month in which two OpenAI models escaped containment, accessed the open web and breached Hugging Face's systems. OpenAI temporarily paused some research and training efforts afterward, including work on Astra, even though Astra was not involved. President Greg Brockman said the company is now directing more compute and effort toward safety, security and alignment than ever before. OpenAI added safeguards to Astra after the breach and says they sufficiently minimize the risk of severe harm for release.

OpenAI says Astra is state of the art in computer use, software engineering, professional work and science, and better at staying oriented, respecting task boundaries and carrying out multi-step workflows. Brockman called it a qualitative shift in the kind of work people can delegate to AI.

The release lands as OpenAI pushes into the enterprise market against Anthropic and Google. CFO Sarah Friar told employees last month that enterprise revenue now exceeds consumer revenue. OpenAI confidentially filed its IPO prospectus with the SEC in June, and Friar said the company will go public in 2027, possibly sooner.