Gemini 4 Carbon in GitHub Copilot is on unless admins opted out

GitHub made Gemini 4 Carbon generally available in Copilot on November 10, 2026, on the Pro+, Max, Business and Enterprise plans. In Business and Enterprise organizations that keep default model enablement on, members get Carbon without any admin action. Its rate is $2 per million input tokens and $10 per million output tokens until December 1, then $4 and $20. Admins now have to decide whether to let Carbon stay on as inherited or set it on purpose before the price change.

Carbon costs the same per token as Gemini 4 Argon today, but Google says Carbon spends more tokens than Argon on the hard problems it targets. A team that drifts from Argon to Carbon can pay more per task now, and Carbon's per-token rate doubles in December.

What changed with Gemini 4 Carbon in GitHub Copilot?

Carbon is now a GA option in the model picker, billed at Google's list price under usage-based billing. The changelog entry describes it as designed for "complex coding, agentic, and reasoning tasks, with a focus on secure code generation and security-focused work." It names VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the coding agent, the Copilot app, github.com and GitHub Mobile. Rollout is gradual, so some users won't see it yet. Copilot Pro is not on the plan list.

Google released Carbon to paying developers and enterprises the same day, two weeks after a first release to cybersecurity defenders. Google positions it above Argon: "Argon remains the ideal choice for the majority of software engineering and knowledge work," while Carbon suits "complex, multi-step coding and debugging." Google also cites benchmark rankings and cost-per-task claims. Those are Google's numbers; the changelog adds no results of its own.

Is Gemini 4 Carbon enabled by default?

Yes, in Business and Enterprise organizations that keep the global default on. The changelog says that "new models are automatically enabled unless an administrator has turned off the global default or explicitly disables this model."

The switch behind this is the Default availability policy. An unconfigured GA release shows up in admin settings labeled Delegate to Default Policy, and the enterprise settings guide says that if this policy is on, such entries "will be available for everyone." Our explainer on the global Copilot model policy shows how inheritance and delegation fit together.

Claude Fable 5.2, launched a day earlier, went the other way: it is excluded from default enablement and stays off until an admin acts, as covered in the Fable 5.2 retention opt-in. Carbon has no such exclusion, so it arrives on wherever Default availability is enabled.

How much does Gemini 4 Carbon cost in Copilot?

Carbon costs $2 per million input tokens and $10 per million output tokens until December 1, 2026, then $4 and $20. The pricing reference converts token spend into AI credits at $0.01 each. Business and Enterprise plans pool each user's included credits per billing entity, and usage beyond that pool is charged at these rates. Google's launch post gives matching dates, with its introductory price ending November 30.

Per 1 million tokens, as listed on November 11, 2026:

Scroll sideways to see all 4 columns.

Option	Input	Cached input	Output
Gemini 4 Carbon, until Dec. 1	$2.00	$0.10	$10.00
Gemini 4 Carbon, from Dec. 1	$4.00	$0.20	$20.00
Gemini 4 Argon, until Dec. 1	$2.00	$0.10	$10.00
Claude Opus 5.5	$4.00	$0.20	$20.00
GPT-6.1 Sol (up to 272K input)	$2.00	$0.10	$10.00
Claude Fable 5.2	$8.00	$0.20	$40.00

Argon's promotional rate ends on the same date, but the reference doesn't say what Argon costs afterward. Opus 5.5, Fable 5.2 and GPT-6.1 Sol also carry a cache write charge that the Gemini rows don't list.

BaristaLabs arithmetic, uncached: one long agent request that reads 200,000 input tokens and writes 20,000 output tokens costs $0.60 (60 AI credits) on Carbon today and $1.20 (120 credits) from December 1. That matches Opus 5.5 at $1.20 and is half of Fable 5.2 at $2.40. Since Google says Carbon uses more tokens than Argon on similar problems, an equal rate per token doesn't mean an equal bill per task. For how pooled allowances work at the plan level, see when bundled Copilot in GitHub Enterprise costs less.

Which teams should use Carbon instead of Argon?

Google's own guidance is Argon for most jobs and Carbon for the hardest ones. Google says Carbon spends extra tokens to "explore thoroughly, reason deeply, and run extended verification loops." BaristaLabs interpretation: that suits long debugging sessions, multi-package migrations and security review, and adds spend without much benefit on a routine edit.

A newly added frontier option can also get picked for jobs that don't need it, simply because it's new. Until December 1, that habit costs the same per token as Argon. After that, each routine request sent to Carbon is billed at twice Argon's current per-token rate, with no change in the task.

One caveat before switching it off: the settings guide warns, "If you disable any of the models required for Fusion Max, then Fusion Max will not be available for use." The pages cited here don't say whether Carbon is among them, so teams on Fusion Max should check first. See setting fusionTier for background.

Why it matters for teams that own Copilot spend

BaristaLabs interpretation: the upside is real. Organization users can get a frontier coding option at a low introductory rate, with no ticket to IT. The risk is that nobody chose it. Something that arrived through inheritance has no owner, no budget line and no review date, and its rate doubles in three weeks.

The cheapest time to make Carbon a deliberate choice is before December 1. Teams can compare it with Argon and Opus 5.5 while Carbon is still at the introductory rate. If usage then exceeds your pooled allowance, the budget conversation starts with data instead of a surprise. Our piece on budget increase requests covers what an approval changes.

Recommended first step: make Carbon an explicit setting before December 1

BaristaLabs recommendation: turn the inherited state into a written decision for this one release.

1. Find the row. In your enterprise, open AI controls, then Copilot, then Configure models, and look for Gemini 4 Carbon. A Delegate to Default Policy label means it is following the inherited setting.
2. Pick a status on purpose. Choose Enabled, Disabled, or delegate it to one pilot organization or enterprise team with hard, long-running jobs.
3. Check Fusion Max if you plan to disable it.
4. Size the December bill. Look at your pooled AI credit headroom and assume Carbon requests cost twice as much per token after November.
5. Decide whether the global switch should stay on. Turning off Default availability makes every unconfigured GA model default to disabled, including ones members already use and future releases, not just Carbon. That is a broader policy change, so give it an owner.

To set an AI policy and a credit budget your teams can live with, see AI consulting.

Sources and scope

• Gemini 4 Carbon in GitHub Copilot, GitHub Changelog, November 10, 2026
• Models and pricing for GitHub Copilot and Managing availability of models in your enterprise, GitHub Docs, retrieved November 11, 2026
• Gemini 4 Carbon is now available, Google, November 10, 2026

BaristaLabs did not run Gemini 4 Carbon while writing this. Prices, dates, plan availability and inheritance behavior come from the pages above as of November 11, 2026. The cost example is arithmetic from published rates, not a measured bill, and does not account for how many tokens Carbon uses on a given task. Lines marked BaristaLabs interpretation, arithmetic or recommendation are BaristaLabs' views, not those of GitHub or Google.