Gemini 4 Carbon API: model ID, pricing and how to get access

The Gemini 4 Carbon API is the part of Google's announcement every developer wants to try. Here is exactly where it stands today, what it will cost, and the code you can run right now so that switching to Carbon takes one line on the day it opens.

Is the Gemini 4 Carbon API available?

Google is taking the same careful path it took with Argon: security teams first, everyone else after. It is also taking part in the US government's voluntary testing of AI models before release. Argon went from announcement to public API in 13 days, but Google hasn't promised the same timing for Carbon, so don't plan a launch around a guess.

So if you call the Gemini API today, Carbon won't be there. You can still do every bit of the setup now.

What is the Gemini 4 Carbon model ID?

Google has not published one. Carbon is not on the Gemini API models page, where the newest model is Gemini 4 Argon, gemini-4-argon. You will see posts guessing at names like gemini-4-carbon or gemini-4-carbon-preview. Don't hard-code a guess: a wrong model name simply fails.

The reliable way to know the moment it lands is to ask the API which models your key can use:

# pip install google-genai
from google import genai

client = genai.Client()  # reads GEMINI_API_KEY
for m in client.models.list():
    if "carbon" in m.name.lower():
        print("Carbon is live for this key:", m.name)

Or with curl:

curl -s "https://generativelanguage.googleapis.com/v1beta/models" \
  -H "x-goog-api-key: $GEMINI_API_KEY" | grep -i carbon

We will add the official model ID to this page as soon as Google publishes it.

Gemini 4 Carbon API pricing

Google did publish the price, and it is the same as Gemini 4 Argon's, introductory discount included:

Google hasn't said when Carbon's intro price ends. For comparison, Argon's pricing page lists its intro price until November 30, 2026, with the standard price from December 1. Try our Gemini 4 Carbon cost calculator to price your own jobs.

What the Gemini 4 Carbon API can do

Google has not yet published Carbon's official context window, output limit, knowledge cutoff or input types. Its long-context results were measured on prompts between 256K and 1M tokens, and Argon, the model it builds on, has a 1,048,576-token window. Treat anything more specific you read elsewhere as a guess until Google's model page says it.

Get ready: code you can run today

The trick is simple: build on Gemini 4 Argon, which is live and costs the same, and keep the model name in one environment variable. On Carbon's launch day you change GEMINI_MODEL and nothing else.

1. Get an API key. Sign in to Google AI Studio and create a Gemini API key. Argon is on the paid tier, so turn on billing. Keep the key out of your code and out of git.
2. Set two environment variables. GEMINI_API_KEY for the key, and GEMINI_MODEL for the model name (today gemini-4-argon).
3. Install the official SDK. pip install google-genai for Python, or npm install @google/genai for JavaScript.
4. Swap the model on launch day. When Carbon's ID appears in the models list, update GEMINI_MODEL and run your tests.

Python

import os
from google import genai
from google.genai import types

client = genai.Client()  # reads GEMINI_API_KEY
MODEL = os.environ.get("GEMINI_MODEL", "gemini-4-argon")

response = client.models.generate_content(
    model=MODEL,
    contents="Review this function and list every security bug you find.",
    config=types.GenerateContentConfig(max_output_tokens=8000),  # cap the bill
)
print(response.text)

JavaScript / Node

import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI({}); // reads GEMINI_API_KEY
const MODEL = process.env.GEMINI_MODEL ?? "gemini-4-argon";

const response = await ai.models.generateContent({
  model: MODEL,
  contents: "Explain what this stack trace means and how to fix it.",
  config: { maxOutputTokens: 8000 },
});
console.log(response.text);

curl (REST)

curl "https://generativelanguage.googleapis.com/v1beta/models/${GEMINI_MODEL:-gemini-4-argon}:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -X POST \
  -d '{"contents":[{"parts":[{"text":"Explain the CAP theorem in three sentences."}]}]}'

OpenAI-compatible endpoint

Already using the OpenAI SDK? The Gemini API has an OpenAI-compatible endpoint, so you only change the key, the base URL and the model name:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["GEMINI_API_KEY"],
    base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
)
reply = client.chat.completions.create(
    model=os.environ.get("GEMINI_MODEL", "gemini-4-argon"),
    messages=[{"role": "user", "content": "Write a haiku about carbon."}],
)
print(reply.choices[0].message.content)

Google hasn't said whether Carbon will be served through this endpoint on day one, so test it when access opens.

Using a model that runs for days

Carbon's headline for developers is stamina: Google says it can keep working on one engineering task for days. A few habits keep long runs useful and affordable:

• Stream long answers. Print output as it arrives instead of waiting for one giant response:

for chunk in client.models.generate_content_stream(model=MODEL, contents=long_task):
    print(chunk.text or "", end="", flush=True)

• Give it a clear finish line. Say what "done" looks like: the failing test passes, the report covers every file. Google says Carbon is better than Argon at asking when a task is unclear, so answer its questions.
• Cache what repeats. Cached input is 95% off at the intro price, which adds up fast when an agent rereads the same codebase all day.
• Cap every call. Set max_output_tokens to what the step needs, and watch the running bill on long agent loops.

Google AI Studio, Vertex AI and other platforms

Google's announcement promises Carbon to developers, businesses and consumers "through our product surfaces" but names no platform or date. Argon is a useful guide to what usually happens: it arrived on the same day in the Gemini API, Google AI Studio, Vertex AI (now called Gemini Enterprise Agent Platform) and the Gemini app for paying subscribers. Expect a similar spread for Carbon, but check Google's docs before you plan around any one platform.

Gemini 4 Carbon API questions

Sources: Google's Carbon announcement, evaluation methodology, Gemini API models, Gemini API pricing, OpenAI compatibility.