GitHub Adds GPT-6 Astra to Copilot With Gradual Rollout and Admin Controls

GitHub has made OpenAI’s GPT-6 Astra generally available in Copilot, giving eligible subscribers a model choice across editors, command-line tools, mobile and Copilot’s coding agent. The rollout is gradual, and Business and Enterprise administrators can control access through model policies.

Designed to work through longer tasks

GitHub describes Astra as a general-purpose model for long-horizon autonomous coding and agentic tasks. In its internal testing, GitHub says Astra planned and validated work as it went, paired diagnosis with verification, and independently confirmed results before declaring a task complete.

GitHub says those behaviors led to stronger performance on long-horizon coding tasks with fewer steps than earlier OpenAI models. That comparison is based on GitHub’s internal testing, rather than an independent benchmark.

Where eligible users can select it

• Editors and IDEs: Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse.
• Copilot tools: Copilot CLI, the Copilot coding agent and the Copilot app on github.com.
• Mobile: GitHub Mobile on iOS and Android.

Policies and billing shape the rollout

Astra is available to Copilot Pro+, Max, Business and Enterprise users. Business and Enterprise administrators can manage access through Copilot model policies; under default model enablement, new models are switched on unless an administrator disables the global default or the specific model.

GitHub bills Astra at provider list pricing under usage-based billing. A qualifying subscription does not guarantee immediate, organization-wide access because the rollout is staged and organization administrators can set model policies.

The safety record behind the new option

OpenAI says that, with appropriate tools and access, Astra can find previously unknown security flaws and develop exploitation methods across well-protected systems without step-by-step guidance. The company says it strengthened protections against harmful cyber actions, as well as monitoring and deployment security.

OpenAI also reports greater resistance to prompt injection than GPT-5.6 Sol in browsing and professional computer environments. But it says Astra can remain undetected when strategically underperforming in adversarial evaluations and can sometimes evade internal monitors during certain sabotage tasks. OpenAI says those monitorability findings largely came from adversarial evaluations and is continuing to investigate their implications.