Gemini 4 Carbon Not Showing in Copilot? Check Plans, Policies and Rollout

Gemini 4 Carbon became generally available in GitHub Copilot on November 10, with a gradual rollout. If it is missing from your model picker, check your Copilot plan and administrator policy before changing your setup. GitHub lists Pro+, Max, Business and Enterprise as eligible plans, not Free or Pro.

The launch announcement covers VS Code, Visual Studio, Copilot CLI, coding agent, the Copilot app, GitHub.com, GitHub Mobile, JetBrains, Xcode and Eclipse. It also explicitly warns that availability will arrive gradually. A release being generally available does not establish that your particular account has received it yet.

First check which account and plan you are using

Start by writing down the GitHub account signed into the affected client and the Copilot plan attached to that account. This sounds basic, but it is more useful than reinstalling software before establishing whether the account is eligible.

A subscription to another company's AI product is not evidence of a Copilot entitlement. Likewise, the word “Pro” is insufficient here: the launch names Pro+ and Max among individual plans. Verify the full plan name instead of assuming every paid subscription includes the same model catalogue.

GitHub's model overview says available models depend on the plan, client, administrator policies and changing model availability. Keep those as separate checks. They explain different failure modes and should not be collapsed into a vague conclusion that the model is broken.

For an initial comparison, inspect the same account on another supported surface, such as GitHub.com. This is a diagnostic suggestion, not a documented guarantee that switching clients accelerates rollout. Record what each surface shows so an administrator or support agent has something concrete to investigate.

Ask an administrator about the effective model policy

For Business and Enterprise, the launch says new models follow default enablement unless an administrator has disabled that default or the specific model. A developer cannot infer the policy from a colleague's screenshot, especially when access rules differ between groups.

GitHub's enterprise model-management guide describes both organization-based access and an opt-in enterprise-team preview. In team mode, organization-level model controls are deactivated. An enterprise-wide disabled model cannot be enabled for a team, while membership in multiple teams can provide access to additional enabled models.

The useful question is therefore: “Which policy applies to my account?” Asking only whether Carbon is enabled somewhere in the company can produce a technically correct answer that does not solve your problem.

An administrator should inspect the relevant enterprise baseline and delegated organization or team setting. Changing a global default to fix one person's access could expose many other models to many other users, so use the narrowest change consistent with the company's approved policy.

Separate a missing model from a usage-budget problem

If Carbon appears but a request cannot run, you have a different symptom from a missing picker entry. Record the exact error and check usage controls before assuming that the launch has not reached you.

The Copilot pricing reference meters model interactions using input, cached input and output tokens, converted into AI credits. One AI credit equals $0.01. Its current Carbon table lists $2 input, $0.10 cached input and $10 output per million tokens; recheck the live table before budgeting future work.

As an illustrative calculation, 100,000 uncached input tokens and 10,000 output tokens at those listed rates total $0.30, or 30 credits. That is not a measured cost for a coding task. An agent session can involve multiple interactions, so treating one prompt as one fixed-price request will underestimate some workloads.

GitHub documents a Bundled AI credits budget with an enforced stopping option. A team should decide that budget deliberately rather than discovering its limits halfway through an important debugging session. Model eligibility and permission to spend are related operational questions, but they are not the same check.

Send a support report that narrows the problem

Once the account, plan and policy checks are complete, preserve a small evidence packet: the affected client and version, the time checked, whether Carbon is absent or produces an error, and whether the same account behaves differently on another supported surface. Include the administrator's policy finding when applicable.

For example: “My Business seat is active. Carbon is missing in VS Code and GitHub.com. Our administrator confirmed the applicable model policy allows it. Checked at 14:00 UTC.” That hypothetical report is much easier to act on than “Carbon does not work.” Do not include repository secrets or access tokens in screenshots.

Because GitHub has not promised a per-account rollout deadline in the announcement, there is no verified waiting period after which every missing entry must be a local fault. Avoid repeatedly changing permissions or subscriptions simply to force an appearance.

Once the model is available, choose one difficult task with a clear acceptance test before making it the team's default. Getting access resolves the setup question; whether Carbon earns its usage cost still needs evidence from the work you actually do.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical teams turn product changes and developer questions into clear, useful content.

Book a 30-minute call.