Gemini 4 Carbon: Google’s Cybersecurity AI Model Explained

Gemini 4 Carbon is Google’s new top model, and you probably cannot use it yet. On 28 October 2026, Google announced Gemini 4 Carbon, which it calls its strongest model with state of the art cybersecurity and long horizon skills. Instead of a public launch, Google is handing it first to a small group of vetted security teams.

Here is what Gemini 4 Carbon does, how it compares with Gemini 4 Argon, what it costs, and why Google is keeping it behind a gate.

Quick answer: Gemini 4 Carbon is a Google DeepMind model built on Gemini 4 Argon and tuned for finding, exploiting and patching software flaws in a controlled way. Right now only vetted cyber defenders get access through the Fairwind Program. Google plans wider release later but has not given a date.

What is Gemini 4 Carbon?

Gemini 4 Carbon is a frontier AI model from Google. Koray Kavukcuoglu, Google’s Chief AI Architect, introduced it on the Google blog. It arrived just two weeks after Gemini 4 Argon and builds directly on it.

The focus is security and long tasks. Google says Carbon can work for days on hard engineering jobs, such as tracing the root cause of a bug across a huge codebase and then writing the fix. That kind of patient, multi step work is exactly what security teams need.

If you missed the earlier model, our guide to Gemini 4 Argon explains the base that Carbon grew from.

What Gemini 4 Carbon can do

Google tested Carbon on the three main jobs of a code security team. Note that Google ran these tests without its usual safety guardrails, and no outside group has repeated them yet.

TaskBenchmarkGoogle’s reported score
Finding vulnerabilitiesSEC-bench Pro88.2%
Exploiting themExploitGym54.8%
Patching themCWE-bench71.0%

Google also makes several broader claims:

  • Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59.
  • It more than doubles Argon’s success rate at writing secure code.
  • The Chrome team saw more than ten times better results than another lab’s large model when generating good patches.
  • It showed a 0.0% attack success rate on the Gray Swan indirect prompt injection test, down from 0.7% for Argon.

In addition, Google reports safety gains over Argon. High severity misalignment alerts in live coding fell by almost 80%, and unsupported claims dropped by about half.

Why Google is limiting access

A model that finds and exploits bugs is a double edged tool. Defenders can use it to fix holes before anyone notices them. However, attackers could use the same power to break in. So Google is giving defenders a head start.

Access goes first to a subset of cyber defenders. Next, it widens to more trusted partners through the Fairwind Program. Only after more feedback and guardrail work will developers, businesses and consumers get it.

This mirrors a wider trend. Anthropic runs a similar gated approach for its most capable model, Claude Mythos Preview, through Project Glasswing. AI leaders keep warning that the same tools that protect systems can also attack them, a theme we covered in our piece on Sam Altman’s AI cybersecurity warning.

Who can get Gemini 4 Carbon now?

The Fairwind page lists who can apply:

  • Governments and national cyber authorities.
  • Critical infrastructure operators in healthcare, telecom, energy and finance.
  • Core technology platforms.
  • Academic labs that focus on defensive benchmarks.

In addition, Fairwind already has more than 650 partners. Named ones include Accenture, CrowdStrike, Datadog, Deloitte, Palo Alto Networks, Snowflake and Wiz.

To apply, an organization contacts its Google Cloud account team. Teams that are not Google Cloud customers start with Google Cloud’s setup docs. Expect vetting and background checks on your security history.

Partners must also follow strict rules. For example, they need phishing resistant multi factor login, access only for internal security, incident response or penetration testing teams, and tracking of each employee’s use. Sharing or reselling access is banned, and so is building malware. Note that, as of 2 November, the Fairwind page still lists Argon, and Google has not published separate written terms for Carbon.

Free: AI Tools Cheat Sheet 2026

24 free tools sorted by task, a simple prompt formula and 10 prompts you can copy. We send the PDF straight to your inbox.

Free. No spam. Unsubscribe in one click.

How much does Gemini 4 Carbon cost?

Google shared pricing in its announcement:

  • Introductory price: $2 per million input tokens and $10 per million output tokens.
  • Cached input: 95% off the input price.
  • After the introductory period: $4 per million input tokens and $20 per million output tokens.

Google has not said which platforms will offer Carbon, such as the Gemini API, Vertex AI or the Gemini app. For now, treat these prices as planning numbers for security teams.

What if you are not eligible?

For now, most developers and companies will wait. Meanwhile, you still have options:

  1. Use CodeMender, Google’s code security agent, with publicly available Gemini models.
  2. Use Gemini 4 Argon where your plan offers it for secure coding help.
  3. Keep basic hygiene strong: patch fast, scan dependencies and review AI written code before you ship it.

Students who want to learn should look at CodeMender on Google Cloud, which the Fairwind page suggests instead of applying.

What Gemini 4 Carbon means for everyone else

Even if you never touch Carbon, it will still affect you. After all, faster bug finding means faster patches for the apps and browsers you use every day. You may see more security updates, and more of them may credit AI tools for the discovery.

At the same time, keep some healthy doubt. Almost every number above comes from Google. Forbes also pointed out that no independent group has reproduced the results. Real proof will come when partners share what Carbon finds in the wild.

Also read: Gemini Skills Replace Gems on 17 November: How to Move Without Losing Work

FAQ

What is Gemini 4 Carbon?

It is Google’s most capable model as of late October 2026, built on Gemini 4 Argon and tuned for cybersecurity and long, multi step tasks.

Can I use Gemini 4 Carbon now?

Not as a regular user. Only vetted cyber defenders get access through the Fairwind Program. Google has not given a public release date.

How do I join the Fairwind Program?

Contact your Google Cloud account team. Eligible groups include governments, critical infrastructure operators, major tech platforms and defensive research labs.

How much does Gemini 4 Carbon cost?

The introductory price is $2 per million input tokens and $10 per million output tokens. Later it rises to $4 and $20.

How is Carbon different from Gemini 4 Argon?

Carbon builds on Argon, scores higher on Google’s security tests, writes secure code more reliably and has narrower access for now.

Leave a Comment

AI Prompts That Work: A Simple 5 Part Formula Best AI Note Taker Apps for Meetings and Lectures AI Agents Passed Secrets Through a CPU Cache: What Happened Samsung TV Privacy: How to Turn Off Tracking Today Copilot+ PC Features: What Only These Laptops Can Do n8n vs Zapier vs Make: Which AI Automation Tool to Pick AI Hallucination: Why AI Makes Things Up and How to Stop It Best AI Browser Right Now: Comet, Chrome, Edge and More Xbox Nano Controller: Price, Size and Who It Is For AI Voice Cloning Scams: How to Spot Them and Stay Safe