How to use Gemini 4 Carbon in GitHub Copilot for agentic coding

GitHub says Gemini 4 Carbon is generally available in Copilot, bringing another model option to developers working on complex, multi-step coding tasks. For readers looking at how to use Gemini 4 Carbon in GitHub Copilot for agentic coding, the announcement sets out where the model is intended to fit and which Copilot experiences it supports.

In its November 10, 2026 announcement, GitHub describes Carbon as suitable for complex coding, reasoning, agentic tasks and security-focused work. Those are GitHub’s descriptions of the model’s intended uses; they do not establish how it performs against other models on a particular project.

Announcement detail	GitHub’s stated scope
Release status	Generally available, with a gradual rollout
Named plans	Copilot Pro+, Max, Business and Enterprise
Example entry points	Supported IDEs, Copilot CLI, coding agent and GitHub.com
Intended workloads	Complex coding, reasoning, agentic tasks and security-focused work

What changed in GitHub Copilot on November 10, 2026?

The GitHub changelog identifies Carbon as generally available across a range of Copilot clients, including Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse. It also names the Copilot app, GitHub Mobile and GitHub.com. The rollout is gradual, so the model may not appear immediately.

Agentic coding describes a workflow in which a developer gives an assistant a task that can involve multiple actions. In an IDE, GitHub says Agent mode can edit files, propose terminal commands and iterate on a task. This makes the model choice part of a broader workflow that includes reviewing the resulting changes.

GitHub’s models and pricing documentation lists Carbon as a generally available model with usage-based token pricing. That distinction matters when planning a task: selecting a model determines more than the assistant’s name. Teams assessing how assistants fit into existing processes may also find a project consultation useful for defining workflow requirements.

Who can access Gemini 4 Carbon for agentic coding?

GitHub’s announcement names Copilot Pro+, Max, Business and Enterprise as eligible plans. Access also depends on the Copilot client and the model options available there, so check the live model picker for the experience you intend to use. GitHub’s supported-models documentation explains that availability varies by plan and client.

Who sees Carbon?

GitHub lists Gemini 4 Carbon as a selectable model for its cloud agent at supported entry points. The cloud-agent model guide also explains that Auto is used when a model picker is unavailable. For a task that depends on using Carbon specifically, check the task’s model selection before assigning the work; the absence of a picker does not confirm that Carbon was selected.

Model access is tied to the Copilot experience as well as the subscription. A developer who can choose Carbon in one supported interface should still check the available options in another before planning a workflow around it. GitHub’s Copilot plans page shows which model options apply across its features.

What administrators control

For Business and Enterprise accounts, administrators can manage model access through policies. If Carbon is missing from an eligible organization account, a developer can ask an administrator to check the relevant model policy and the organization’s Copilot configuration.

This distinction matters when a team standardizes an agentic workflow: an individual’s model picker reflects the options available to that account, while organization policy can affect what the wider team can select. Teams defining how AI assistants fit into existing processes can consider those access decisions alongside the workflow integration and review practices described in Techno Particles’ generative AI services.

How to select Gemini 4 Carbon in an IDE Agent session

In Visual Studio Code, start with a bounded coding task, such as adding validation to an existing form and updating the associated tests. Define the expected behavior and the files or component involved so the agent has a clear scope for its work.

Select Agent mode and the model

Open Copilot Chat and select Agent mode. Then use the current-model dropdown in the chat interface to choose Gemini 4 Carbon. GitHub’s model-selection guide describes the dropdown workflow. The Chat model choice does not change the model used for inline suggestions.

Write the request in terms of the result you want: for the form example, identify the inputs that need validation, describe the messages users should see, and ask for tests covering valid and invalid submissions. If the repository has established conventions, point the agent to the relevant files so it can work within them.

For a mobile project, a similarly bounded task might involve validating one screen’s inputs rather than changing the entire sign-in flow. The scope can be recorded alongside the requirements in an application development plan, making the intended behavior clear to the people reviewing the change.

Review edits and commands

GitHub’s Agent mode guide explains that the agent can edit files, propose terminal commands and iterate. Inspect the proposed changes against the requested behavior, including any tests it adds or modifies. Approve a terminal command only after checking what it will do; reject a command that does not fit the task.

If the result misses a requirement, give the agent a specific correction, such as identifying an input case that its validation does not handle. Review the resulting changes before accepting them into the project.

How to use Gemini 4 Carbon through Copilot CLI

Copilot CLI offers a terminal-based way to select a model for a coding session. This is useful when the work already starts in a repository’s command-line environment and the task can be described as a concrete change to that project.

Choose a model interactively

Start a Copilot CLI session and enter /model to open the available model list. Select Gemini 4 Carbon if it appears. GitHub’s CLI model-selection guide describes this interactive workflow and how the selection persists for the session.

After selecting the model, give the agent a task with a defined outcome. For example, ask it to update a command-line application’s argument handling so that a missing required option produces a clear error. Identify the relevant command and expected behavior, then ask for coverage of the missing-option case.

A task description can also name the boundaries of the change: the agent should work on argument handling without altering unrelated command behavior. Those boundaries give the resulting work a clear purpose and make the request easier to repeat in a later session.

Pin a model for a prompt

The CLI also supports specifying a model when starting a prompt. GitHub documents this generic command:

copilot --model MODEL-NAME -p 'YOUR PROMPT'

Replace the placeholders with the identifier shown in your available model list and the task you want to run. The example does not establish a particular identifier for Carbon, so use the value offered by your own Copilot experience. A recorded prompt and model choice can help a team understand how a task was initiated.

For a website repository, the task can be as focused as adding an accessible label to one existing form field. The implementation context may differ across projects; the website development service page describes the broader types of web work in which such a change can occur.

Credits, token pricing, and cost control

GitHub lists separate rates for input tokens, cached input tokens and output tokens for Gemini 4 Carbon. It also identifies AI Credits as the consumption mechanism for Copilot agentic features. Both details matter when estimating the cost of a multi-step coding task: the model’s rates describe the usage being billed, while the Copilot plan determines the credit context in which the work runs.

Promotional token rates

According to GitHub’s model pricing documentation, Carbon’s promotional rates through December 1, 2026 are $2.00 per million input tokens, $0.10 per million cached input tokens and $10.00 per million output tokens. The listed rates then become $4.00, $0.20 and $20.00 per million tokens, respectively.

For a simple illustration, one million input tokens and 100,000 output tokens would correspond to $3.00 at the promotional rates, before considering cached input or plan-specific billing. The calculation is $2.00 for input plus $1.00 for output. It is an illustration of the published rates, not a prediction of the cost of a particular coding session.

AI Credits in agentic workflows

GitHub’s Copilot plans page identifies Agent mode, coding agent, Copilot CLI and Copilot Chat as features that use AI Credits. Model options vary by feature, so a team should check its applicable plan and billing controls when forecasting usage.

Actual spend depends on the amount of input, cached input and output, the model used and the controls attached to the plan. For a team planning a broader rollout, documenting the intended workflows and budget constraints can make that forecast more useful; Techno Particles’ project consultation services describe scope and planning support for software projects.

When to choose Gemini 4 Carbon

Choose Gemini 4 Carbon when a coding task needs multi-step reasoning and that workflow justifies its credit and token cost. To put how to use Gemini 4 Carbon in GitHub Copilot for agentic coding into practice, select it for one bounded task and assess whether the result meets your requirements.