# Gemini 4 Carbon: Google’s new top model goes to cyber defenders first

URL: https://ai-pulse.aglaganova.com/blog/gemini-4-carbon-googles-new-top-model-goes-to-cyber-defenders-first
Published: 2026-10-29
Author: Roman Aglaganov, Full-stack developer
Category: Models
Region: Global
Tags: safety, coding

> Google says Gemini 4 Carbon is its strongest model, built for long engineering jobs and for finding and fixing security flaws. Only selected cyber defenders can use it for now, while Google takes part in voluntary US government testing.

![Illustration: a black hexagonal lattice like a carbon sheet, with a thin grey shield outline over it and one red atom where a bond is broken](https://ai-pulse.aglaganova.com/api/media/file/gemini-carbon-1.png)

Google announced Gemini 4 Carbon on 28 October, two weeks after Gemini 4 Argon. It calls Carbon its strongest model, with what it describes as state-of-the-art abilities in cybersecurity and in “long horizon” tasks that run for hours or days. Most people cannot use it yet.

## Who gets it

Google is starting with a subset of cyber defenders, then gradually adding trusted partners through its Fairwind Program, the vetted-access scheme it uses for security models. It says it is taking part in the US government’s voluntary process for testing models before release, and will make Carbon available to developers, businesses and consumers “as soon as possible” after gathering feedback and refining guardrails. There is no date yet. The introductory price is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 later.

## What Google claims

All figures here are Google’s own. It says Carbon scores 88.2% on SEC-bench Pro, a test of finding real vulnerabilities; 54.8% on ExploitGym, which tests whether a model can show that a flaw can actually be exploited; and 71.0% on CWE-bench, which tests writing working fixes. Carbon ranks first, at 59, on version 5.0 of the Artificial Analysis Intelligence Index, an independent combined benchmark, ahead of Anthropic’s Claude Opus 5.5 at 58 and OpenAI’s GPT-6 Astra at 57, according to The Hacker News.

Google says its own teams have seen Carbon work for days on hard engineering problems, such as tracking down and patching subtle bugs across large codebases. It says Chrome’s team found Carbon more than ten times better than “another lab’s large frontier model” at producing high-quality patches, without naming the rival.

## Why the slow rollout

A model that is good at finding and exploiting flaws helps defenders, but it could help attackers too. That is why Google, like other labs, is giving its strongest security models to vetted defenders first. Google says Carbon builds on Argon’s safety measures and adds alignment training for complex, high-pressure tasks. In its internal tests, it says, high-severity “misalignment alerts” in live coding work dropped by almost 80%, Carbon made about half as many unsupported claims as Argon, and it resisted every indirect prompt-injection attack in one outside benchmark. Those results come from Google and still need checking by others.

## Sources

- [Google: Gemini 4 Carbon, our next frontier model](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)
- [The Hacker News: Google’s Gemini 4 Carbon leads benchmarks for finding, exploiting, and patching vulnerabilities](https://thehackernews.com/2026/10/googles-gemini-4-carbon-leads.html)