As of November 12, 2026, Google has announced API access to Gemini 4 Carbon for paying developers and enterprises, and is rolling out product access to Google AI Ultra subscribers. Its Gemini Developer API model ID is gemini-4-carbon; the model is not available on the free tier. A product entry point or paid status does not guarantee availability for a particular account, region or project, so check your access in the relevant product before relying on it.
We reviewed Google's availability announcement, the model and pricing documentation, and the Fairwind page without signing in, applying or calling Carbon. This guide reflects public information, not a test of an account's access.
Choose your next step
| Your situation | Google's announced entry point | Next step |
|---|---|---|
| API developer | Carbon is available through the paid Gemini API tier. The model directory lists gemini-4-carbon; the free tier does not offer it. |
Check your region, project billing and quota, then consult the Carbon documentation and your account's actual access. |
| Google AI Ultra subscriber | Carbon is rolling out to Ultra subscribers in the Gemini app, Google Antigravity and AI Mode in Search. | Check the model selector and account notifications in your product; rollout and region can affect availability. |
| Enterprise user | The announcement lists Gemini for business via API or model selector, as well as Antigravity and CodeMender. | Confirm your enterprise account's access and the applicable product, region and costs before use. |
| Cyber defense organization | Carbon for cyber is rolling out through Fairwind to eligible trusted partners. | Contact your Cloud account team as described on the Fairwind page; applying does not guarantee approval or immediate access. |
Google AI Ultra subscriptions and the Gemini Developer API paid tier are separate products and billing paths. A subscription fee does not replace API token charges, and an API key or product entry point alone does not establish that your account has Carbon access.
As of November 19, 2026, Google AI Pro users with personal Google Accounts can select Hydrogen or Neon in the Gemini app’s Chat and Task modes. Ultra adds Carbon. Start with Hydrogen for routine work, choose Neon for deeper analysis, or consider Carbon for advanced coding and long-horizon tasks. Availability depends on your plan and remaining usage. Gemini Apps model availability · Compare Neon, Carbon and Hydrogen · AI platform subscription and API pricing
What should API developers check first?
The Carbon model documentation lists gemini-4-carbon as a stable version. Before integrating, check that the Gemini API is available in your region and that your project has the appropriate paid billing setup and quota. Confirm actual account access: a model listing alone does not guarantee it. Google's billing and rate limits documentation explains the prerequisites.
The documented context window is 1,048,576 tokens shared between input and output. The default synchronous output limit is 65,536 tokens, including thinking and output tokens; longer output requires long decode. Carbon supports implicit context caching only, not explicit caching. When migrating an application, consult the model documentation and Carbon migration guide rather than assuming another model's parameters and caching behavior apply unchanged.
What does the Carbon API cost?
At the time checked, the Gemini Developer API pricing page listed these Standard paid-tier rates in US dollars per million tokens:
| Usage | Through November 30, 2026 | Starting December 1, 2026 |
|---|---|---|
| Input | $2 | $4 |
| Output, including thinking tokens | $10 | $20 |
| Implicitly cached input | $0.10 | $0.20 |
For example, one million ordinary input tokens plus one million output tokens costs $12 at the introductory Standard rates, or $24 at the rates starting December 1. This simple estimate assumes the output total includes thinking tokens and excludes caching, tools, other charges and taxes. Actual charges depend on usage and applicable terms.
The caching rate applies only to input that actually hits the implicit cache; a hit is not guaranteed on every request. Explicit caching is unsupported for this model. Google also lists Batch, Flex and Priority options at different rates. Do not apply this table automatically to other platforms or Carbon for cyber; check the platform, service option and current prices before use. Google's announcement specifies that its post-introductory prices apply to the Carbon model intended for public access.
How is Fairwind different from public Carbon?
Carbon for cyber is a separate variant for defensive cybersecurity work. Google Cloud documentation says it is restricted by the Fairwind allowlist; the public Carbon API ID or paid access does not confer access to that variant.
The Fairwind page says it prioritizes governments and partners critical to societal resilience, including critical infrastructure operators and core technology platforms, and vets applicants. Access is controlled, and partner organizations may not share, resell or redistribute access to the models.
Organizations interested in applying should read the eligibility and governance conditions and contact their Cloud account team. Organizations that are not Cloud customers can start with the Google Cloud getting-started information linked from the page. Google does not promise a specific review timeframe; an application or existing Fairwind membership does not guarantee Carbon for cyber access. Refer to Google's response for the decision.