---
title: "Gemini 4 Carbon: Google's Frontier Model Built to Find and Fix Vulnerabilities"
paper: "The Daily Model"
edition: 2026-10-29
newsroom: Google
edition_id: tip-20261029-799bf328-approved
content_hash: 75008c9b43a32775d8784907e32f237a2fc9cbe7e570f5b66d9a721cc76e8828
sources:
  - Gemini 4 Carbon: our next frontier model (https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/) 2026-10-28
---

# Gemini 4 Carbon: Google's Frontier Model Built to Find and Fix Vulnerabilities

## The short version

- **Can I use Gemini 4 Carbon right now, and how do I get access?** Carbon is available only to a subset of cyber defenders starting this week, with gradual expansion through the Fairwind Program before broader access.
- **Are the cybersecurity benchmark scores independently verified?** All benchmark figures (SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0%) are reported by Google in its announcement; no independent verification is included.

A vulnerability sits in production code for months. Nobody finds it because nobody has the hours. Then a model reads the codebase, names the flaw, proves it can be exploited, and writes the patch. That is the sequence Google says Gemini 4 Carbon now leads on three separate cybersecurity benchmarks.

Google announced [Gemini 4 Carbon on October 28, 2026](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/), calling it their strongest model with state-of-the-art cybersecurity and long-horizon capabilities. The benchmark numbers are the headline: 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for validating exploitability, and 71.0% on Collinear's CWE-bench for turning findings into working code fixes. Google notes these external benchmarks were evaluated without guardrails—a qualification that matters when considering how the scores translate to deployed use. Google also reports Carbon ranks #1 on Artificial Analysis Intelligence Index v5.0 at a score of 59. All of these figures are Google-reported; the announcement does not provide independent verification.

What makes the cybersecurity story concrete is the internal evidence Google cites. On internal benchmarks, Google says Carbon achieves more than 2x uplift in secure-code generation over its predecessor Argon. The Chrome team reportedly observed Carbon performing more than 10X better than an unnamed rival lab's frontier model when generating high-quality patches. Wiz, Google's cloud-security platform, observes Carbon leading on web penetration-testing benchmarks tied to its Scan for Good initiative. The unnamed competitor in the Chrome comparison makes that particular claim hard to evaluate from the outside.

The rollout is deliberately narrow. Starting the week of October 28, Google is making Carbon available to a subset of cyber defenders, expanding gradually through what it calls the Fairwind Program before opening access to developers, businesses, and consumers. Google says it is participating in the U.S. government's voluntary pre-release model-testing process. Pricing matches Argon's introductory rate: $2 per million input tokens, $10 per million output.

On the safety side, Google reports Carbon hit a 0.0% attack-success rate on Gray Swan's Indirect Prompt Injection Benchmark, improving on Argon's 0.7%. Google also says high-severity misalignment alerts in live internal coding dropped by almost 80%, and unsupported claims were cut roughly 2x versus Argon.

The question for defenders is whether benchmark leadership translates into fewer missed vulnerabilities in real codebases under time pressure. Google's answer so far is a controlled cohort and a promise of wider access soon. The rest of us wait for the Fairwind Program to widen.