Gemini 4 Carbon is available to everyone: price, access and first test

👁 112 reads

Two weeks ago Gemini 4 Carbon was a model you could only read about: Google published a table of records and gave access to the US government and a few hundred security teams in its Fairwind program. On 10 November 2026 Google AI Studio announced that Carbon “is here”. An hour later I had it running in my own scripts and put it through the same tasks I used for Gemini 4 Argon in October.

The short version: Carbon costs exactly the same as Argon in the API, it is available in the European Union, but there is no free tier and it is not included in Google AI Pro.

Where you can use it

• Gemini API in Google AI Studio - for all paying developers. Model ID: gemini-4-carbon. The Gemini API changelog lists it as generally available (stable) from 10 November.
• Gemini for business - for companies, via the API.
• Google AI Ultra - Carbon is rolling out in the Gemini app, AI Mode in Search and Google Antigravity. AI Mode with Carbon is limited to “select countries”, without a list.
• Google Antigravity - Gemini 4 Carbon appears in the model picker with Low, Medium and High thinking levels (High by default), for Ultra and Enterprise only, not for Free, Plus or Pro. The app also shows a “Turbo” switch that the documentation does not describe yet.
• OpenRouter - google/gemini-4-carbon, same price as Google. This is how I ran my tests.
• Also: Android Studio (with a paid Gemini API key), CodeMender for enterprises, and a program for scientists who can apply for six months of free access.

Price: 2 and 10 dollars, but only until the end of November

The Gemini API price list shows introductory rates for Carbon: 2 USD per million input tokens and 10 USD per million output tokens (thinking tokens included), valid until 30 November 2026. From 1 December the price doubles to 4 and 20 USD.

The context window is 1,048,576 tokens, shared between input and output. By default the model returns up to 65,536 tokens. Caching is implicit only. There is no free tier for Carbon in the API - the price list says so explicitly. In the Gemini app Carbon is part of Google AI Ultra: $99.99 (5x) or $199.99 (20x) a month. Google AI Pro ($19.99) gives you Argon, not Carbon.

How to start

Open Google AI Studio, enable billing on your project, create a key and pick Gemini 4 Carbon. The thinking level is set with thinking_level (low, medium or high, medium by default).

A warning from my first hour: Carbon thinks for a long time. With a 6,000-token output limit it spent 5,808 tokens on reasoning and returned an empty answer - and I still paid for it. Google’s documentation gives the same advice: control cost with the thinking level, not with a low output limit.

First results from my test

• 23 practical tasks (agent with tools, code with unit tests, writing and proofreading, puzzles, a 158,000-character document): all correct, apart from one point my grader wrongly deducted, exactly as with Argon. Cost 0.23 USD, time 264 seconds. Argon did the same in 135 seconds for 0.22 USD.
• Made-up sources: asked for 8 scientific papers with DOIs on two topics, Carbon listed 16 publications that all exist, with two minor errors (one wrong DOI, one wrong year). Argon had invented 2 of 16 in October.
• Security review of a small Flask app: all 8 planted vulnerabilities found with correct lines and CWE numbers, in 67 seconds for 0.03 USD.

For everyday tasks Carbon does nothing Argon cannot do, and it does it more slowly. Google itself says Argon remains the right choice for most software engineering and knowledge work. Carbon is meant for long, hard tasks that need searching, checking and fixing. All prices, tests and comparisons are in my Gemini 4 Carbon guide, and the full test with Claude Fable 5.2 and GPT-6.1 Sol is in my Gemini 4 Carbon test.

// read moreGemini 4 Carbon test: 50+ tasks vs Argon, Claude Fable 5.2 and GPT-6.1 SolAI ToolsGemini 4 Carbon: Benchmarks, Price and AccessAI ToolsGemini 4 Carbon benchmarks: all 19 tests and where it losesBenchmarks

Kinetyra