Independent guide · Updated Oct 29, 2026

Gemini 4 Carbon

Google just announced Gemini 4 Carbon, its strongest AI model yet, two weeks after Gemini 4 Argon went public. Here is everything that is actually confirmed: the benchmarks, the price, the API and when you can use it.

Announced Oct 28, 2026
Public API: not yet
#1 on 15 of 17 tests

Gemini 4 Carbon API See the benchmarks

Not affiliated with Google. Every number below comes from Google's announcement or named press, linked at the bottom.

Gemini 4 Carbon at a glance

The short version, as of October 29, 2026.

AnnouncedOct 28, 2026On Google's blog, by Koray Kavukcuoglu of Google DeepMind

Built forCoding, agents, cyber defenseLong jobs that can run for days

BenchmarksFirst on 15 of 17In Google's table vs GPT-6 Astra and Claude

Price (intro)$2 in / $10 outPer million tokens. Later $4 / $20

API statusNot public yetNo model ID published so far

Who gets it firstCyber defendersEveryone else "as soon as possible"

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's newest and strongest AI model. Google DeepMind announced it on October 28, 2026, and calls it a frontier model with state-of-the-art cybersecurity skills and the stamina for very long jobs. It is the second model of the Gemini 4 generation, after Gemini 4 Argon, which was announced on September 30 and opened to developers on October 13.

You may see it written as "Gemini Carbon 4". The official name is Gemini 4 Carbon: the generation number comes first, then the model's name, just like Gemini 4 Argon.

Three things stand out in the announcement:

It beats Argon everywhere. Carbon scores higher than Argon on all 17 tests in Google's table, and it tops GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 on 15 of them.
It is built for cyber defense. It finds, exploits and patches software vulnerabilities better than any model Google compared it with, which is why security teams get it first.
It works for a long time. Google says Carbon can run for days on big engineering jobs, like tracking down and fixing a bug across a large codebase, and that it asks questions when a task is unclear instead of guessing.

Want the details? Read when you can get it, what it costs, or what it does for security teams.

Gemini 4 Carbon benchmarks

Google's own numbers, from its announcement table.

All benchmarks →

Terminal-Bench 4.0

Real work in a computer terminal, scored by Artificial Analysis. Higher is better.

Gemini 4 Carbon65.2%

Claude Opus 5.559.6%

GPT-6 Astra59.1%

Gemini 4 Argon57.1%

Across Google's full table of 17 tests, Carbon is first on 15, second on one and third on one. See where it loses.

AA Intelligence Index59Ranked #1 on version 5.0

SEC-bench Pro88.2%Finding real security bugs. Next best: 85.4%

ExploitGym54.8%Proving a bug is exploitable. Next best: 42.4%

GraphWalks96.2%Reasoning over 256K to 1M tokens of text

For developers

Gemini 4 Carbon API

Here is the straight answer: the Gemini 4 Carbon API is not open to the public yet. Google is giving Carbon to a small group of cyber defenders first, then to more trusted partners through its Fairwind Program, and says developers, businesses and consumers come next "as soon as possible". There is no model ID yet, and Carbon is not on the Gemini API model list.

Input$2 / 1MIntro price. Standard: $4

Output$10 / 1MIntro price. Standard: $20

Cached input95% offAbout $0.10 per 1M at the intro price

The good news: Carbon launches at exactly the same price as Gemini 4 Argon, and Argon is live in the Gemini API today as gemini-4-argon. Build on Argon now, keep the model name in one setting, and moving to Carbon is a one-line change:

# pip install google-genai
import os
from google import genai

client = genai.Client()  # reads GEMINI_API_KEY from your environment
MODEL = os.environ.get("GEMINI_MODEL", "gemini-4-argon")  # swap in Carbon's ID when Google publishes it

response = client.models.generate_content(model=MODEL, contents="Find the root cause of this failing test.")
print(response.text)

Read the full Gemini 4 Carbon API guide for how to spot Carbon the moment your key gets it, JavaScript and curl examples, and pricing math.

While you wait for Carbon

Make videos and images with Google's best models today

Carbon is a thinking, coding and security model: it writes text and code, it doesn't make pictures or video. Google's video makers are Veo 3.1 and Gemini Omni, and its top image model is Nano Banana 2. You can use all three right now on Lumeta, next to Seedance, Kling and more, in one account.

Every top image, video, voice and music model in one place
One account and one set of credits, instead of five subscriptions
Made to cost less than Higgsfield, Magnific and Artlist

Try Lumeta

Full disclosure: Lumeta is made by the team behind this site.

Veo 3.1Google's flagship video model, with sound Gemini OmniCreate and edit videos by chatting Nano Banana 2Google's image maker and editor Seedance 2.5Cinematic video from text or images Kling O3Video with control over every character Upscale VideoSharpen any clip up to 4K

Gemini 4 Carbon guides

Short, sourced answers to the questions people are searching right now.

DevelopersGemini 4 Carbon APIModel ID status, pricing and code you can run today. ScoresBenchmarks vs GPT-6 and ClaudeAll 17 results, including the two Carbon doesn't win. CostPricing and cost calculator$2 / $10 per million tokens, and what a real job costs. AccessRelease date and who gets itCyber defenders, Fairwind, then everyone else. CompareGemini 4 Carbon vs ArgonWhat changed in two weeks, test by test. SecurityCarbon for cyber defenseFinding, exploiting and patching bugs, and the Fairwind Program. TodayWhat to use while you waitThe best options while Carbon rolls out. AnswersGemini 4 Carbon FAQEvery common question, answered in a sentence or two.

Gemini 4 Carbon FAQ

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's newest and strongest AI model, announced on October 28, 2026. Google built it for coding, long-running agent work and cyber defense, and it beats Gemini 4 Argon on every test in Google's benchmark table.

Is it Gemini 4 Carbon or Gemini Carbon 4?

The official name is Gemini 4 Carbon. "Gemini Carbon 4" is a common way people type it, and it means the same model.

When is the Gemini 4 Carbon release date?

It was announced on October 28, 2026, and is rolling out in stages. A small group of cyber defenders gets it first, then more trusted partners through Google's Fairwind Program. Google says developers, businesses and consumers come next "as soon as possible", with no exact date yet.

Is the Gemini 4 Carbon API available?

Not publicly yet. Google has published the price ($2 per million input tokens and $10 per million output tokens at launch) but not a model ID. Gemini 4 Argon, at the same price, is available in the Gemini API today.

Can Gemini 4 Carbon make videos or images?

Google has not said it does. Carbon is a text, code and security model. Google's video makers are Veo 3.1 and Gemini Omni, and its image model is Nano Banana 2, which you can use today on platforms like Lumeta.

See all Gemini 4 Carbon questions →

Sources: Google's announcement, Google's evaluation methodology, 9to5Google, Gemini API models.