OpenAI’s GPT-6 Astra is built for long, difficult tasks that require more than one good answer. The model targets coding, computer use, research, and document work where it must keep context, use tools, and check whether it actually finished the job. For builders, test whether Astra’s persistence reduces the supervision a real project needs.

OpenAI positions Astra as its most capable model yet. For a real multi-step task, compare the corrections, restarts and manual checking it needs with your usual model.

What changes for builders

OpenAI’s announcement describes Astra as a model for long, multi-step work rather than quick one-off answers. In GitHub’s own testing, Astra stood out for planning and checking its work as it went. It also confirmed its results before declaring a task done. GitHub says that led to stronger performance on long-horizon coding tasks with fewer steps than prior OpenAI models.

That points to a specific workflow change. Astra may be better suited to tasks such as tracing a bug across several files, making a change, running checks, and revising the result. It does not mean you can stop reviewing its work. A model that can act for longer can also make a larger incorrect change before you notice it.

Availability is spreading through existing tools

GitHub announced that GPT-6 Astra is available to Copilot Pro+, Max, Business, and Enterprise users. You can select it in VS Code, Visual Studio, Copilot CLI, the Copilot coding agent, the Copilot app, github.com, GitHub Mobile, JetBrains IDEs, Xcode, and Eclipse. GitHub says the rollout is gradual, so an eligible account may not see it immediately.

That Copilot rollout matters because it puts a frontier model inside a tool many builders already use. It also gives the launch a practical constraint: GitHub says Astra is billed at provider list pricing under usage-based billing. In other words, access through Copilot does not make the model’s underlying usage cost disappear.

What it costs

OpenAI’s pricing page lists Astra at $10 per million input tokens and $50 per million output tokens. Tokens are small pieces of text a model reads or generates. A request with more than 272,000 input tokens, roughly a large codebase context, costs more again under OpenAI’s long-context pricing rules.

For a reference point, GPT-5.6 Sol has a promotional rate of $4 per million input tokens and $20 per million output tokens through at least November 21, 2026. Astra is therefore 2.5 times Sol’s current promotional price per input or output token. That is a price comparison, not a claim that Sol is every builder’s default model.

Administrators control model access

If you’re on a Business or Enterprise plan, GitHub’s changelog says admins manage access through the model policy in Copilot settings. New models are enabled for everyone by default unless an administrator has turned off the global default or explicitly disabled Astra.

If you manage a team, check that policy before the rollout reaches everyone. The risk is not that Astra is automatically used for every request. The risk is that people can choose a much more expensive model without understanding how it affects usage.

What builders should do next

Don’t decide based on benchmark claims. Pick a coding task you already run regularly, such as tracing a bug across three files without changing unrelated behavior. Run it once on your usual model and once on Astra. Compare the number of corrections, the time to a working result, the usage consumed, and the fixes required after review.

Use Astra where the extra persistence solves a problem you actually have. Keep a cheaper model for routine edits if Astra produces similar results at a higher usage cost.

End of article