Google released Gemini 4 Carbon to developers on November 10, 2026. The model runs via paid access to the Gemini API in Google AI Studio and Android Studio, and businesses get access via Gemini for business. For Google AI Ultra subscribers, Google is enabling Carbon in the Gemini app, in AI Mode in Google Search, and in the Antigravity development environment.
Previously, only selected cyber defense specialists were allowed to use the model, about this first phase TutKit reported at the end of October. For most Gemini users, the release changes nothing: Carbon is missing from the Google AI Plus and Google AI Pro subscriptions, and Google does not offer a free tier in the API.
Does the release also apply in Germany?
Google mentions no exception for Germany or the EU and speaks of all Ultra subscribers; a list of countries is missing. Gemini API and AI Studio are available in Germany, Austria, and Switzerland, and AI Mode in Search runs in all three countries.
How do you get Gemini 4 Carbon?
Access since November 10, 2026, as Google describes it
- Google AI Ultra: Gemini app, AI Mode in Search, Antigravity
- Paid access to the Gemini API in AI Studio and Android Studio
- Businesses via Gemini for business
- Google AI Plus and Google AI Pro subscriptions
- Free tier of the Gemini API
Google is enabling the model for Ultra gradually.
Sources: Google Blog, Google AI Studio Docs, Google One
According to the announcement, Google is enabling Carbon gradually for Ultra subscribers. If the model is missing from the model selection in your Gemini app, your account may not have its turn yet. In the plan overview from Google One, Carbon appears only with Ultra, while Pro includes Gemini 4 Argon.
What can Carbon do better than Argon?
Carbon builds on Argon and is intended for extensive tasks with many work steps: programming, web development, data analysis. For office work, Google promises that Carbon consolidates results from extensive documents and spreadsheets and, when creating spreadsheets, presentations, and reports, adheres strictly to specified structures and formats.
A proprietary Google test with 375 tasks checks whether a model admits knowledge gaps instead of inventing information. Carbon solved 92 percent of them correctly, Argon 70 percent. These figures have not been independently verified.
According to the developer documentation, input, reasoning steps, and response together may comprise at most 1,048,576 tokens. Tokens are the text building blocks by which Google bills. Carbon outputs responses over 65,536 tokens only if you let the request run in the background and pick up the result later. According to the model's data sheet, its knowledge cutoff extends to March 2026.
What does Carbon cost until the end of November?
Google states prices only in dollars and without a euro price. Until November 30, 2026, the same introductory price applies to Carbon as to Argon. From December 1, according to the pricing page, the prices for input and output double, for Carbon as well as for Argon. Switching to Argon therefore saves nothing on the price per token from December.
A single call can nevertheless cost more with Carbon than with Argon: the model uses more tokens per response for reasoning steps, and Google bills these intermediate steps as output. Jobs that do not need to be finished immediately cost half the standard price in the batch and flex tiers.
What this means for you
With Google AI Plus or Pro, you don't need to do anything; Carbon is not included there. What distinguishes the free version of Gemini from the paid one is shown in the first lesson of the Gemini crash course. If you have Ultra, it's best to test Carbon with a real task, such as a report from several PDFs and spreadsheets. Still, compare the figures in the result with your source documents, because Google itself concedes that the model can invent information.
If you use the API for your own scripts or client projects, try out Carbon at the introductory price until November 30. When switching from Argon, according to Google, the new model ID gemini-4-carbon is sufficient. However, increase the limit for output and the time your program waits for responses, because Carbon takes longer than Argon. For quotes from December on, it's best to calculate with doubled token prices in dollars until Google names a euro price.
Google released Gemini 4 Carbon on November 10, 2026.
Sources
