Google Launches Gemini 4 Carbon on GitHub Copilot and Vercel AI Gateway

Google released Gemini 4 Carbon on November 10, 2026, bringing its latest coding, reasoning, and cybersecurity model to the Google Developer API, GitHub Copilot, and Vercel’s AI Gateway. While the model is now generally available across all three ecosystems, access rules, rollout speed, and pricing differ significantly depending on where you use it.

How to access Gemini 4 Carbon and what it costs

Before switching your model picker or API key, here is how availability and billing actually work across each platform based on their November 10 release notes:

Platform	Who Can Use It & Rollout Status	Supported Tools & Surfaces	Confirmed Pricing & Billing
GitHub Copilot	Copilot Pro+, Max, Business, and Enterprise users (not available on Copilot Free or standard Copilot Pro). Gradual rollout—check back soon if it has not appeared in your picker yet.	VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, Copilot coding agent, Copilot app, github.com, and GitHub Mobile (iOS/Android)	Billed at provider list pricing under usage-based billing
Vercel AI Gateway	Live now for all Vercel AI Gateway accounts using model ID google/gemini-4-carbon	Vercel AI SDK, OpenAI-compatible Chat Completions API, Claude Code, Codex, Cursor, and fx via npx vercel ai-gateway setup	50% off through November 30, 2026; zero platform fee and no markup on provider pricing (including on Bring Your Own Key / BYOK requests)
Google Developer Platform	Live now for developers via Google AI Studio and Gemini API	Direct REST/SDK calls and multimodal agent pipelines	Standard metered Google developer API token pricing

GitHub Copilot: Gradual rollout, eligible plans, and admin controls

In its November 10 changelog, GitHub confirmed that Gemini 4 Carbon is entering general availability for Copilot Pro+, Max, Business, and Enterprise subscribers, but cautioned that rollout will be gradual. If you are on an eligible plan and do not yet see Gemini 4 Carbon in your model picker across Visual Studio Code, JetBrains IDEs, Xcode, Eclipse, Visual Studio, github.com, GitHub Mobile, or Copilot CLI, it is still propagating to your account.

Two practical details matter for anyone watching their monthly bill or managing a team:

• Usage-based billing at provider list pricing: Using Gemini 4 Carbon inside GitHub Copilot is billed under GitHub’s usage-based billing model at Google’s provider list pricing rather than an unlimited flat-rate seat allowance.
• Default enablement for Business and Enterprise: On Copilot Business and Copilot Enterprise plans, administrators control access via the model policy in Copilot settings. Under GitHub’s default model enablement setting, Gemini 4 Carbon is enabled automatically unless an admin has turned off the global default or explicitly disabled this model.

Vercel AI Gateway: 50% off through November 30 and 1M-token context

For independent developers, startups, and product teams calling models via API or coding agents like Cursor, Claude Code, and Codex, Vercel’s announcement includes the most immediate cost benefit: Gemini 4 Carbon is 50% off on Vercel AI Gateway through November 30, 2026.

Beyond the promotional discount, Vercel confirmed that AI Gateway passes through provider pricing with no markup and no platform fee on inference, including when teams use Bring Your Own Key (BYOK). Developers can connect external coding agents in one step by running npx vercel ai-gateway setup and selecting google/gemini-4-carbon.

What Gemini 4 Carbon actually improves

Both Google and Vercel highlight that Gemini 4 Carbon ships with a 1-million-token context window tailored for large codebases and long documents. According to the release documentation, the model focuses on three practical areas:

• Multi-stage software engineering: Following through on complex full-stack web development and multi-file refactoring while respecting project constraints.
• Cybersecurity defense: Locating software vulnerabilities, validating whether a risk is real, and drafting working patches.
• Fewer unsupported claims: Google reports more reliable execution of multi-step plans and better collaboration when user prompts or product requirements are ambiguous, tested alongside Google’s new Multimodal Knowledge Agent benchmark.