The ChatGPT GPT Astra rollout has begun through a limited release, pairing the new model’s expanded capabilities with additional safety monitoring. OpenAI’s release notes say Astra improves coding, research, computer use and complex multi-step work, while access is initially limited to a set of organizations.

The release notes, updated on Sept. 3, say Astra can create documents, spreadsheets and presentations that follow a user’s templates and instructions. OpenAI also says the model can adapt when requirements change during a task. The same notice says broader availability is planned over the following days, but it does not describe a full public launch.

The rollout is connected to a wider safety program. In a Sept. 1 safety update, OpenAI said its evaluations placed Astra above the company’s critical cybersecurity capability threshold. The company said the model could find previously unknown security flaws and develop exploitation methods across well-protected systems when given the right tools and access.

That assessment is why the release is not being described as a normal model update. OpenAI says stronger safeguards are required during development and before release. The company’s safety material discusses alignment evaluations, abuse prevention and monitoring intended to identify cases in which an agent may act outside an authorized scope.

Axios reported on Sept. 8 that OpenAI is also using Astra in work on difficult mathematical and theoretical computer-science problems. That report describes the company’s claimed results and places the model within a broader push toward longer-running AI systems. The report does not change OpenAI’s stated access limits.

For ChatGPT users, the practical point is that the name Astra does not automatically mean the model is available to every account. OpenAI’s own release note says rollout is limited and that wider availability is planned. Access, features and safeguards may therefore differ by organization and product.

The confirmed update is both a model release and a deployment decision. OpenAI is presenting Astra as a more capable system for research, coding and computer use, but it is also keeping the initial distribution narrow while additional monitoring is applied.

Claims about unrestricted access, autonomous operation or a final public release should not be treated as confirmed unless OpenAI updates its official documentation. The current evidence supports a limited rollout with new capabilities and an unusually prominent safety process.