Gemini 4 Carbon FAQ

Gemini 4 Carbon is Google's newest and strongest AI model, announced by Google DeepMind on October 28, 2026. It is built for coding, long-running agent work and cyber defense, and Google says it is state of the art at cybersecurity.

The official name is Gemini 4 Carbon: the generation (4) comes first, then the model name (Carbon), like Gemini 4 Argon. "Gemini Carbon 4" is a common way people search for it and means the same model.

Google DeepMind. The announcement was written by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google.

Google hasn't explained the name. Like Argon before it, it is a chemical element, which press coverage describes as Google's new naming scheme for Gemini 4 models.

Carbon is the newer, stronger model. It beats Argon on all 17 tests in Google's benchmark table at the same price, makes unsupported claims about half as often, and is better at asking when a task is unclear. See Carbon vs Argon.

It was announced on October 28, 2026. Selected cyber defenders get it first, starting that week, then more partners through Google's Fairwind Program. Developers, businesses and consumers come "as soon as possible", with no date yet.

Only if you are a vetted security team in Google's Fairwind Program. Everyone else can use Gemini 4 Argon today, which costs the same.

Not yet. Google says consumers will get it after defenders and trusted partners. Argon is in the Gemini app for Google AI Pro and Ultra subscribers.

Google has not announced a free version. Argon, priced the same, needs a paid API account or a Google AI Pro or Ultra plan.

Fill in the interest form at deepmind.google/fairwind-program. It is for governments, critical infrastructure operators, major tech platforms and academic security labs, and Google vets every applicant.

Not publicly yet. Google has announced the price but not opened access or published a model ID. See our API guide for code that switches to Carbon in one line.

Google hasn't published one. Don't trust guesses like gemini-4-carbon until it appears on the Gemini API models list.

$2 per million input tokens and $10 per million output tokens at launch, with cached input 95% off. After the introductory period: $4 and $20. That is the same as Argon.

Google hasn't published it. Its long-context test used prompts of 256K to 1M tokens, and Argon's window is 1,048,576 tokens.

On Google's own table it comes first on 15 of 17 tests, and Artificial Analysis ranks it first on its Intelligence Index v5.0 with a score of 59. Claude Opus 5.5 still leads on operating a computer, and GPT-6 Astra on one hard coding test.

On most of Google's tests, yes. It beats GPT-6 Astra on 16 of 17 and Claude Opus 5.5 on 14 of 16. See all benchmarks.

Yes. It leads Terminal-Bench 4.0 (65.2%) and SWE-sweep, climbed to fourth on WebDev Arena, and Google says it can work for days on large engineering tasks. GPT-6 Astra and Claude Opus 5.5 still edge it on FrontierSWE 2.

Google says it uses the same layered safety system as Argon, resisted every prompt-injection attack in Gray Swan's test, and is going through the US government's voluntary pre-release testing. Its strong hacking skills are why access is limited to defenders for now.

Google hasn't said it can. Carbon writes text and code. For video, Google's models are Veo 3.1 and Gemini Omni; for images, Nano Banana 2. All three are on Lumeta today, made by the team behind this site.

No. GeminiCarbon4.com is an independent guide, not affiliated with Google. See About.