Playbooks · Teams
Test Gemini 4 Carbon in Copilot before Dec 1
Written November 15, 2026
Carbon is GA in Copilot and its rates double December 1. The cheap test window, the thinking-token catch, and the December 14 forced swap behind it.
Carbon is generally available in Copilot
GitHub's November 10 changelog post announces that "Gemini 4 Carbon, Google's latest model, is now generally available in GitHub Copilot," designed "for complex coding, agentic, and reasoning tasks, with a focus on secure code generation and security-focused work." It is available to Copilot Pro+, Max, Business, and Enterprise users, selectable in the model picker across Visual Studio Code, Visual Studio, Copilot CLI, the Copilot coding agent, the Copilot app, github.com, GitHub Mobile, JetBrains IDEs, Xcode, and Eclipse — with the caveat that "rollout will be gradual," so a missing picker entry today is not a verdict. The post also fixes the billing frame: the model "is billed at provider list pricing under usage-based billing," which is what makes the December date below a price change rather than a plan change.
The December 1 step, in GitHub's figures
GitHub's models and pricing reference for Copilot — the page the changelog post points to for billing detail — lists Gemini 4 Carbon at $2.00 per million input tokens, $0.10 per million cached input tokens, and $10.00 per million output tokens, and its footnote states the terms exactly: Carbon "is available at the promotional pricing of $2.00 per 1M input tokens, $0.10 per 1M cached input tokens, and $10.00 per 1M output tokens until December 1, 2026. Starting December 1, 2026, pricing is $4.00 per 1M input tokens, $0.20 per 1M cached input tokens, and $20.00 per 1M output tokens." Every component doubles on the same day. Gemini 4 Argon sits on the identical promotional rates until December 1 under the neighbouring footnote, so a Carbon-versus-Argon comparison run this month is priced on a level field — and the same comparison run in December costs twice as much on the Carbon side.
How to run the test so it means something
Run the evaluation now, on real work, and score it per completed task rather than per token. A rate card comparison alone cannot settle a reasoning model: two models on the same card can consume very different token counts on the same job, and the cheaper-looking model is only cheaper if the finished task costs less. Pick a small set of representative tasks — a hard debugging session, a multi-file change, a security review — run each on Carbon at the promotional rates, record the total cost and whether the result held up, and keep the numbers. That record is the decision asset: after December 1 the same evidence costs double to gather, and it stays valid because the model is not changing, only its price. Admins on Business and Enterprise plans control exposure in the meantime: the changelog post notes that administrators manage access through the model policy in Copilot settings, and that under default model enablement new models are automatically enabled unless the global default is off or the model is explicitly disabled — so check the policy before assuming nobody can spend on Carbon yet.
December 14 is behind the price step
The test window has a second wall. GitHub's November 13 deprecation post lists Gemini 4 Argon among the models deprecated across all Copilot experiences on December 14, 2026, with Gemini 4 Carbon as the suggested alternative — and under default model enablement, the alternatives are automatically enabled unless an administrator has turned the global default off. In other words, Argon users are moving to Carbon on December 14 whether they evaluated it or not; the only choice on offer is whether the evaluation happens at promotional rates before December 1 or at full rates after. The full retirement list and its other swaps are covered in Copilot retires six more models on December 14. Google's own API pricing for Carbon is a separate surface, covered in Gemini 4 Carbon is here: $2 in, $10 out, and is not repeated here.
What to do
Before December 1: confirm Carbon appears in your picker or enable it through the model policy, run the per-task comparison while both models sit on promotional rates, and write down cost per finished task. After December 1, the same test still works — it just costs twice as much to learn the same thing, two weeks before Argon's retirement makes the answer mandatory.
Related playbooks
- Copilot Agent Merge: watch a PR and fix its blockers
- The Copilot app now takes GHES accounts
- Copilot Automations learn from real outcomes
- Copilot CLI signs in with Microsoft Entra
Sources
- Gemini 4 Carbon in GitHub Copilot — GitHub Changelog
- Models and pricing for GitHub Copilot — GitHub Docs
- Upcoming deprecation of selected GitHub Copilot models in December — GitHub Changelog
Back to: Teams playbooks hub · Also: Gemini 4 Carbon is here: $2 in, $10 out · Copilot retires six more models on December 14