OpenAI has introduced GPT-6 Astra, describing it as its most intelligent and aligned model to date.
The model brings together advances in pre-training, reinforcement learning and alignment, with stronger performance across computer use, browsing, software engineering, cybersecurity, science and professional work.
OpenAI said Astra reaches state-of-the-art results on several internal and external benchmarks, including FrontierMath Tier 4, ARC-AGI-3 and ExploitBench.
The company is rolling out the model first to limited enterprise users, with broader access planned for ChatGPT Plus, Pro, Business and Enterprise users, as well as the OpenAI API, Microsoft Azure and AWS Bedrock.
The launch comes with heightened safety attention. OpenAI’s safety overview says Astra is the first model it has broadly deployed to reach the Critical level for cybersecurity capability under its Preparedness Framework.
OpenAI said it also created a new evaluation informed by a recent Hugging Face incident, testing whether a model would go beyond its intended scope when faced with a difficult or impossible task.
According to the company, GPT-6 Astra showed major improvements over GPT-5.6 Sol in that evaluation, with fewer cases of acting outside authorized boundaries.
For enterprises, the launch signals a new phase of frontier AI adoption. Astra is positioned not only as a more capable assistant, but as a model that can handle complex work across software, research, security and business workflows.
The bigger test will be whether OpenAI can scale access while maintaining the safeguards needed for a model with stronger autonomous and cybersecurity capabilities.
