Gemini 4 Carbon: Can You Use It? What to Run Now (2026)

You can't use Gemini 4 Carbon yet unless you are a vetted cyber defender. Google announced it on October 28, 2026 and said it will start with a subset of cyber defenders this week, widening to more trusted partners through its Fairwind Program. Developers, businesses and consumers get it "as soon as possible", with no date given. Until then, the closest model you can actually run is Gemini 4 Argon, which Google launched two weeks earlier, and on Zemith it sits next to Claude Opus 5.5 in the same chat on the Professional plan.

TL;DR

• Gemini 4 Carbon was announced on October 28, 2026. Access starts with a small group of cyber defenders and phases in through Google's Fairwind Program. There is no public date for developers, businesses or consumers.
• Google says Carbon will launch at the same introductory API price as Argon: $2 per million input tokens and $10 per million output tokens, then $4 and $20 after the introductory period. It hasn't said how long that period lasts.
• Google's headline claims are about security work: finding, exploiting and patching vulnerabilities. Those are Google's own numbers, measured without guardrails, and nobody outside the program has verified them.
• Gemini 4 Argon is available now to developers and in the Gemini app for Google AI Pro and Ultra.
• Zemith doesn't have Carbon. It has Argon and Claude Opus 5.5 on the Professional plan ($29.99 a month), and the plan to test them on your own work is below.

What is Gemini 4 Carbon?

Google's announcement calls Carbon "our strongest model with state-of-the-art cybersecurity and long horizon capabilities", building on Gemini 4 Argon. The post leads with security results: 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploiting them) and 71.0% on CWE-bench (patching them). It also says Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 at 59, and moves up on WebDev Arena.

Google says the external benchmarks were run without guardrails. That matters because the version a developer or consumer gets later will have guardrails on, and Google says it is still iterating on them before wider release. So the published numbers describe the restricted version, not necessarily the one you will be able to buy.

Can I use Gemini 4 Carbon today?

Only if you are in the program. Google's post says access starts "this week" for a subset of cyber defenders and expands gradually to trusted partners through the Fairwind Program. If you are a security team at a large organization, the route is to ask through your Google Cloud account team. Everyone else waits.

Google's post doesn't mention the Gemini app, Google AI Pro or Google AI Ultra, so don't assume a subscription gets you in. The same goes for Gemini 4 Carbon pricing in consumer plans: only API prices are stated.

How much will Gemini 4 Carbon cost?

For the API, Google says Carbon will launch at $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off the input price. After the introductory period, $4 and $20 apply. That is the same pair of prices as Argon, and the same as Claude Opus 5.5's list price of $4 and $20 from Anthropic's pricing page. Some third-party pages list Carbon at $4 and $20 today, which is the post-intro price, so check Google's API pricing page when Carbon opens up.

Is Gemini 4 Carbon better than Claude Opus 5.5?

Nobody can say yet. Google's comparisons are against its own Argon, and Carbon isn't open for independent runs. For the job Carbon is pitched at, vulnerability research, you can't test it. For everyday coding, writing and analysis, the useful comparison is the one you can run: Argon against Opus 5.5 on your own files, which we set out in Gemini 4 Argon vs Claude Opus 5.5.

What should I use until Carbon is available?

Use Argon if you want to see how Google's current line performs, and Opus 5.5 as the baseline. This is a twenty minute check you can run without Zemith, using any app that has both models:

1. Pick a real security-minded coding job. For example, paste a function you wrote this week. Prompt: "Review this code for security problems. List each issue with the line, why it is exploitable, and a fix. If you are not sure, say so instead of guessing."
2. Add a job with a known answer. Take a bug you already fixed and give the model the old version. See whether it finds the cause you found.
3. Add a long-input job. Paste a long file or a diff and ask: "List anything that changes behavior, not just formatting."
4. Score each answer from 1 to 5 on three questions: Is it correct? Did it invent anything? Could I act on it without rework? Run each job in a fresh chat per model so one answer can't influence the other.

When Carbon opens up, run the same three jobs again and you'll have a baseline to compare against.

Never paste secrets, API keys or customer data into any model while testing. Use a sanitized copy.

How do I run Argon and Opus 5.5 side by side in Zemith?

On the Professional plan, both are in the model picker:

1. Open Zemith and tap Select model.
2. Choose Claude 5.5 Opus and send your first test prompt.
3. Switch to Gemini 4 Argon in the same chat and send the same prompt.
4. Score the two answers, then repeat with your other jobs.

Cost is in credits. A prompt of about 2,000 tokens with an 800-token answer is about 4,900 credits on Argon, and about 4,800 on Opus 5.5, before any thinking tokens. Professional includes 3 million credits a month, roughly 600 exchanges of that size. See the multi-model chat page for the cheaper models you can switch to for routine jobs, and the Zemith page for developers for how people use it for code review.

Should I wait for Carbon before paying for anything?

No. The introductory price Google gave for Carbon is the same as Argon's, and there is no date, so waiting means doing nothing for an unknown number of weeks. If you already pay for Google AI Pro or Claude Pro, you can use Argon or Opus on those. If you want both models in one place for one price, that is the Zemith case, and you can cancel it later. Our Google AI Plus vs Pro post covers what Google's own plans include.

What we don't know yet

As of October 31, 2026, we don't know when Carbon reaches the Gemini app or the public API, what guardrails the public version will have, how long the $2 and $10 introductory price lasts, or how Carbon performs against Opus 5.5 outside Google's chosen benchmarks. We will not add Carbon to Zemith's list until it is available to build on, and we haven't promised a date for it.

Bottom line

Carbon is real, announced and restricted. It is a security-first release that most people can't touch yet. If your work is general coding, writing or analysis, run Argon and Opus 5.5 on your own files this week, keep your scores, and repeat the test the day Carbon opens up.

How we compared: we read Google's Gemini 4 Carbon announcement and its API pricing page on October 31, 2026, and checked Zemith's plans, model list and credit costs in the product's code.