Is Gemini 4 Carbon available? Release date, access and price

Yes. Gemini 4 Carbon became available on November 10, 2026 to every developer on the paid Gemini API tier, to businesses on Gemini for business, and to Google AI Ultra subscribers in the Gemini app. The model ID is gemini-4-carbon. It is not on the free tier.

If you searched this a week ago, the answer was no, and many pages still say so. They were written for the first announcement on October 28, when Carbon went only to a group of cybersecurity defenders. Below is what changed, where you can use it now, and what it costs before and after the introductory price ends.

Gemini 4 Carbon release timeline

In October Google said it was "continuing to gather feedback and iterate on guardrails" before a wider release. That took two weeks. The November post opens with "Today, we're making Gemini 4 Carbon available to developers, enterprises, scientists, and consumers."

What Gemini 4 Carbon is

Carbon is the second Gemini 4 model, after Gemini 4 Argon. Google calls it its strongest model and aims it at long, multi-step work: hard debugging across a codebase, analysis of dense PDFs and spreadsheets, scientific reasoning and security review. It reads text, images, audio, video and PDFs and writes text. Input, thinking and output share one window of 1,048,576 tokens. A single response stops at 65,536 tokens when streaming, but with continuation requests one answer can run to roughly a million tokens. Thinking has three levels, low, medium and high, with medium as the default. Google's own description of the trade-off is that Carbon spends more thinking per turn than Argon, which costs latency and tokens in exchange for more checking.

Who can use Gemini 4 Carbon, and where

Google has not said whether Carbon will come to the Pro or Plus plans of the Gemini app. Today it is an Ultra feature there. The Cloud note about select regions is worth checking before you plan on Vertex-style deployment; the Gemini API route has no such note.

How much does Gemini 4 Carbon cost?

Google's list prices per 1M tokens, checked on November 11, 2026:

That is the same introductory price as Gemini 4 Argon. Two details change real bills more than the headline rates. Thinking is billed as output, and Carbon is the Gemini 4 model that thinks more per turn. Grounding with Google Search is billed per query on top of tokens. Our current per-token rates, read live from billing settings:

Is Gemini 4 Carbon free?

No. Google's pricing page lists Carbon as "Not available" on the free tier. You can try it in the Gemini app only with an AI Ultra subscription. The exception is the science program, which gives accepted researchers Argon and Carbon in Antigravity free for six months.

Calling Gemini 4 Carbon through the API

We called it in production on November 11 through SeedRouter's native Gemini endpoint. The smallest possible check:

curl "https://api.seedrouter.ai/v1beta/models/gemini-4-carbon:generateContent" \
  -H "Authorization: Bearer $SEEDROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"role":"user","parts":[{"text":"Reply with exactly: CARBON_OK"}]}],"generationConfig":{"thinkingConfig":{"thinkingLevel":"low"},"maxOutputTokens":256}}'

It answered CARBON_OK in 3.9 seconds with finishReason: STOP, 38 input tokens and 7 output tokens, and reported standard as the service tier. Over the next hour we sent 40 more requests to check the official capabilities one by one. Thinking levels, structured JSON output, function calling, code execution, Google Search, Google Maps, image, PDF, video and streaming audio input all worked. The request format is the same as Gemini 4 Argon's; Google's guide says switching needs only the model ID.

The API reference has every parameter, and the Gemini 4 Carbon page has a Playground where you can run your own prompts.

Limits we ran into

Long requests should stream. Carbon can think for minutes on hard prompts. One of our non-streaming requests, an audio file, ended with HTTP 524 at the edge after about two minutes. The same request with streamGenerateContent finished in 84 seconds. Google itself recommends client timeouts of at least 15 minutes.

Some settings are refused. minimal thinking, candidateCount: 2, presence and frequency penalties, and a conversation that ends with a model turn all came back as HTTP 400 from the Gemini API. Sampling settings such as temperature were accepted and, per Google, ignored.

URL context ran but could not read pages. The tool executed in both of our tests, and Carbon reported both pages as unreachable, including one of Google's own documentation pages. Paste the text in when you need it read.

Explicit caching is not available. Google lists only implicit caching for Carbon, which applies on its own from 4,096 prompt tokens. Don't plan on cachedContents resources.

Gemini 4 Carbon FAQ

When was Gemini 4 Carbon released?

Google announced it on October 28, 2026 for selected cyber defenders, and released it to paid Gemini API developers, businesses and AI Ultra subscribers on November 10, 2026.

Is Gemini 4 Carbon available in the Gemini app?

Yes, for Google AI Ultra subscribers, along with AI Mode in Search and Antigravity. Google has not announced it for other plans.

What is the Gemini 4 Carbon model ID?

gemini-4-carbon. It is a stable ID, with no preview suffix.

Is Gemini 4 Carbon on the free tier of the Gemini API?

No. It requires a paid tier. Every token it processes is billed, including thinking.

Does the Gemini 4 Carbon price go up?

Yes. Google's introductory rates of $2 input and $10 output per 1M tokens end on November 30, 2026, and $4 and $20 apply from December 1.

Should I use Gemini 4 Carbon or Gemini 4 Argon?

Argon is faster and cheaper per answer for most coding and knowledge work; Carbon is for the hardest problems. We compared them on the same prompts in Gemini 4 Carbon vs Argon.