What GitHub announced
GitHub says both models are designed for agentic coding and code review, with stronger coding judgment than earlier GPT-6 models. The two models have different roles:
- GPT-6.2 Astra: OpenAI's latest Astra model for agentic coding and terminal work. In early testing, GitHub says it caught significantly more code-review issues than GPT-6 Astra and finished agentic coding tasks using noticeably fewer tokens.
- GPT-6.2 Sol: OpenAI's latest Sol model for complex agentic coding and in-depth code review. GitHub says it caught nearly twice as many real code-review issues as GPT-6.1 Sol while keeping its findings accurate.
How much to trust the numbers
These are GitHub's early-testing claims. The announcement gives no benchmark names, sample sizes, test methods or issue counts. It doesn't define "accurate" either, and it offers no precision or false-positive figures. "Nearly twice as many" is a relative comparison. It doesn't say how many real issues the baseline found.
Treat the claims as a reason to run your own trial. They are not a verdict. A useful trial would run the new model and your current default over the same set of pull requests, then count real catches, noise and review time.
Availability and clients
Both models are available to Copilot Pro+, Max, Business and Enterprise users. GitHub's announcement lists the places where you can pick them:
- Visual Studio Code and Visual Studio
- Copilot CLI and the GitHub Copilot coding agent
- The GitHub Copilot app and github.com
- GitHub Mobile on iOS and Android
- JetBrains IDEs, Xcode and Eclipse
Rollout is gradual, so the models may not show up for every eligible user yet. GitHub's advice is to check back later.
GitHub's supported-models documentation lists both models as GA and warns that availability depends on plan and client and may change. Its minimum-IDE-version table lists GPT-6.2 Astra and Sol as "TBD" for every client, so the minimum versions are not yet published. GitHub's general guidance is to keep your IDE and Copilot extension updated.
Pricing: fewer tokens is not a guaranteed lower bill
The models are billed at provider list pricing under usage-based billing, and GitHub gives no per-token rates in the announcement. Using fewer tokens per task should help in principle. But the real cost depends on your usage and on the per-token rates, and GitHub provides no worked cost comparison.
One third-party GitHub pull request, from a project that added the models to its own catalog, listed OpenAI list rates for Astra and Sol. I can't verify those figures against OpenAI or GitHub. Check GitHub's Copilot models and pricing page for current Copilot billing.
What administrators need to do
For Business and Enterprise plans, GitHub says plan administrators can manage access to GPT-6.2 models through the model policy in Copilot settings. Under default model enablement, new models are turned on automatically unless an administrator has disabled the global default or explicitly disabled these models.
If your organization relies on the default policy, developers may see these models without any review having happened. A practical checklist:
- Open the Copilot model policy for your organization or enterprise and see how GPT-6.2 Astra and Sol are set.
- Decide whether to enable them explicitly, or restrict them until your own review is done.
- If users report the models missing, first allow for the gradual rollout. Then check the policy and the user's plan tier.
- Tell developers that generated code still needs human review. GitHub's general guidance says to validate suggestions, including for security, before using them in production. That guidance is not specific to GPT-6.2.
Why it matters
Model churn in Copilot is now fast. GitHub's documentation also shows retirements, including several models retired on November 16, 2026. Teams that pin workflows to a specific model should expect to revisit those choices regularly.
The pairing is also useful. Astra is pitched at terminal-heavy agentic work and Sol at deeper review of complex changes. GitHub doesn't claim either is better overall, so match each model to a task and measure the results yourself.
Bottom line
GPT-6.2 Astra and Sol are live in Copilot for Pro+, Max, Business and Enterprise users, with a gradual rollout. GitHub reports better code-review catches and lower token use, but it hasn't published the data behind those claims. Administrators should check their model policy, and developers should compare the models on their own code before changing defaults.
References
- OpenAI’s GPT-6.2 Astra and GPT-6.2 Sol now available GitHub Changelog · 2026-11-17T20:20:29+00:00
- Supported AI models in GitHub Copilot - GitHub Docs docs.github.com