Google opens Gemini 4 Carbon to API developers and AI Ultra at $2/$10
Google made Gemini 4 Carbon broadly available on November 10, 2026, less than two weeks after launching the model to a small group of cyber defenders. Developers can now call it through the paid tier of the Gemini API in Google AI Studio and Android Studio, Google AI Ultra subscribers can try it in AI Studio and Google Antigravity, and GitHub added it the same day to GitHub Copilot for Pro+, Max, Business and Enterprise users.
The model keeps the price Google announced at launch: $2 per million input tokens and $10 per million output tokens during an introductory period, with cached input tokens at 95% off the input price, followed by $4 per million input tokens and $20 per million output tokens once that period ends. Google describes Carbon as its strongest reasoning model, aimed at multi-step coding, long-horizon agents, scientific research and security work, while it continues to recommend Gemini 4 Argon for most everyday software engineering and knowledge work.
The wider release turns Carbon from a gated security model into a product that a developer with a billing account, a Google AI Ultra subscriber or a GitHub Copilot customer can select today. In the Gemini API the model is listed under the code gemini-4-carbon with a context window of 1 million tokens and text, image, audio, video and PDF input.
··········
GEMINI 4 CARBON AVAILABILITY BY PLAN AND PLATFORM.
Google's developer announcement limits API access to the paid tier, so free-tier keys in AI Studio cannot call the model. Google AI Ultra subscribers get a separate route to try it in AI Studio and in Antigravity, Google's agentic coding environment. In its consumer announcement, Google said Ultra subscribers can also use Carbon in the Gemini app and in AI Mode in Search, while Gemini 4 Argon remains the model for Google AI Pro subscribers and above.
For companies, Google said Carbon is offered through Gemini for business, either through the API or from the model menu, and inside specialized agents such as CodeMender and AlphaEvolve. Google Cloud's documentation for the Gemini Enterprise Agent Platform now includes a Gemini 4 Carbon model page and developer guide, which also cover a cybersecurity variant called Gemini 4 Carbon for cyber.
........
Where | Who can use Gemini 4 Carbon | Status on November 10 |
Gemini API (AI Studio, Android Studio) | Developers on the paid tier | Available |
Google AI Studio and Antigravity | Google AI Ultra subscribers | Available to try |
Gemini app and AI Mode in Search | Google AI Ultra subscribers | Rolling out, per Google |
Gemini for business, CodeMender, AlphaEvolve | Business customers | Available, per Google |
GitHub Copilot | Pro+, Max, Business, Enterprise | Generally available, gradual rollout |
Fairwind Program | Approved cyber defenders | By application |
........
Google also opened two application programs. Scientists can apply for six months of free access to Argon and Carbon through Antigravity's highest usage tier, subject to eligibility, capacity and selection, and Google said 10,000 researchers will receive that access. Cybersecurity defenders who need the specialized variant can keep applying through the Fairwind Program, the channel through which Carbon first reached a subset of security teams at the end of October. Google has not published Gemini app usage limits for Carbon or a country list for the consumer rollout.
··········
GEMINI 4 CARBON IN GITHUB COPILOT.
GitHub's changelog entry dated November 10 lists Gemini 4 Carbon as generally available in Copilot Pro+, Max, Business and Enterprise. Copilot Free and Copilot Pro are absent from that list. GitHub bills the model at provider list pricing under usage-based billing and does not give a request multiplier in the announcement, pointing users to its models and pricing documentation instead.
The model appears in the picker in Visual Studio Code, Visual Studio, Copilot CLI, the Copilot coding agent, the GitHub Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode and Eclipse. The rollout is gradual, so some accounts will see it later than others. On Business and Enterprise plans, administrators control access through the model policy in Copilot settings; with default model enablement, new models are switched on automatically unless an administrator has turned off the global default or disabled Carbon specifically. Teams that want to review Carbon before developers use it should check that policy now. The same path brought Anthropic's newest model to Copilot a day earlier, as covered in our report on Claude Fable 5.2 in GitHub Copilot.
··········
GEMINI 4 CARBON API PRICING COMPARED WITH GEMINI 4 ARGON.
Carbon's per-token prices come from Google's October 28 launch announcement, which set the introductory rate and the later rate but gave no end date for the introductory period. As of November 11, 2026, the official Gemini Developer API pricing page lists Gemini 4 Argon with prices that apply through November 30, 2026 and double from December 1, 2026, and it does not yet carry a Carbon entry. Carbon's list prices match Argon's at both stages, but Google has not confirmed on its pricing page that Carbon's introductory rate ends on the same date. Batch, Flex and Priority prices for Carbon have not been published.
........
Gemini API paid tier, USD per 1M tokens | Gemini 4 Carbon | Gemini 4 Argon |
Input, introductory | $2.00 | $2.00 (through Nov 30) |
Output, introductory | $10.00 | $10.00 (through Nov 30) |
Cached input, introductory | 95% off input ($0.10) | $0.10 |
Input, after introductory period | $4.00 | $4.00 (from Dec 1) |
Output, after introductory period | $20.00 | $20.00 (from Dec 1) |
Batch input / output, introductory | Not published | $1.00 / $5.00 |
Free tier | Not available | Not available |
........
Prices as listed on the official Gemini Developer API pricing page and in Google's Gemini 4 Carbon launch announcement on November 11, 2026.
Equal token prices do not mean equal bills. Google's model page describes Argon as the more efficient model, with faster responses and lower token spend per turn than Carbon, so the same task will usually cost more on Carbon because it reasons for longer and produces more tokens. Google's own advice to keep Argon for routine software engineering follows from that difference. Our earlier Gemini 4 Carbon vs Gemini 4 Argon comparison covers the benchmark gap between the two models in detail.
··········
WHAT A GEMINI 4 CARBON CODING AGENT COSTS BEFORE AND AFTER THE INTRODUCTORY RATE.
A practical way to read the price is to model an agent workload. Assume a coding agent that runs 1,000 tasks a month, with each task sending 300,000 input tokens across its turns and producing 30,000 output tokens, including thinking tokens. Assume that 80% of the input is repeated context served from the cache at 95% off, and that the same 95% cache discount continues after the introductory period, which Google has not stated explicitly for the later rate.
At the introductory price, each task costs $0.12 for 60,000 uncached input tokens, $0.024 for 240,000 cached tokens and $0.30 for output, a total of $0.444. The monthly bill for 1,000 tasks comes to $444. At $4 and $20 per million tokens, with cached input at $0.20, the same task costs $0.888 and the month comes to $888. Without any cache hits the figures would be $900 a month at the introductory rate and $1,800 afterwards.
........
Coding agent, 1,000 tasks a month | Introductory rate | After introductory rate |
Cost per task, 80% cache hits | $0.444 | $0.888 |
Monthly cost, 80% cache hits | $444 | $888 |
Monthly cost, no cache hits | $900 | $1,800 |
........
Output tokens make up about two thirds of the cached scenario, which means caching cuts the bill roughly in half but leaves the output price as the main cost driver. Workloads that send long documents and ask for short answers gain the most from caching, while agents that write long patches or reports will feel the doubling most directly. Developers who want to benchmark Carbon on real tasks at the lower rate have a short window before the price change takes effect, and the broader picture of Google's model prices is in our guide to Gemini API pricing in 2026.
··········
GEMINI 4 CARBON BENCHMARKS AND THE NEW MKA EVALUATION.
At launch, Google reported that Carbon scored 59 on the Artificial Analysis Intelligence Index v5.0, ranking first, along with 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, 71.0% on CWE-bench for patching and fourth place on WebDev Arena with 1768 Elo. These are vendor-reported figures, and several of the security benchmarks use setups that differ between labs, so direct comparison with other models needs care.
On November 10 Google added a new data point with the Multimodal Knowledge Agent benchmark, built with Artificial Analysis. It contains 200 private tasks across 10 verticals and 83 occupations, each modelled on a project that takes a professional three to eight hours and graded against an expert rubric of roughly 70 yes or no checks weighted 75% toward factual accuracy and 25% toward visual design. Google says Carbon ranks second in quality on this benchmark while completing tasks at about six times lower average cost than the top-scoring model, and it publishes 10 public tasks on Hugging Face. The post does not give numeric scores for the models in the text. You can read our coverage of the original restricted release in the Gemini 4 Carbon launch report.
··········
GEMINI 4 CARBON GENERAL AVAILABILITY: CONFIRMED FACTS AND OPEN QUESTIONS.
The confirmed part of the release is clear. Carbon is callable as gemini-4-carbon on the paid Gemini API tier with a 1 million token context window, Google AI Ultra subscribers can use it in AI Studio and Antigravity, and GitHub has made it generally available in Copilot Pro+, Max, Business and Enterprise at provider list pricing. Its token prices are the same as Argon's, at $2 and $10 per million during the introductory period and $4 and $20 afterwards.
Several details are still missing from Google's official pages. The pricing page has no Carbon entry yet, so the end date of the introductory rate, the Batch, Flex and Priority prices and the cache price after the increase are unconfirmed for this model. Google has not published Gemini app limits for Ultra subscribers or a regional list for the consumer rollout. For teams choosing between the two Gemini 4 models, the practical rule from Google's own guidance holds: Argon for most coding and knowledge work, and Carbon for the long, difficult tasks where its extra reasoning tokens are worth paying for, ideally tested before the rate doubles.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
