Gemini 4 Carbon: Cyber-Defense Model, Limited Rollout

Introduction

On October 28, 2026, Google announced Gemini 4 Carbon, which it calls "our strongest model with state-of-the-art cybersecurity and long horizon capabilities." The headline detail is the rollout: Carbon is not generally available. According to Google's post, the model starts this week with "a subset of cyber defenders," and access will phase out to additional trusted partners through the Fairwind Program. Developers, businesses and consumers are promised access "as soon as possible," with no date given.

The Verge headlined its coverage "Google launches Gemini 4 Carbon," but Google's own text describes a gated release. Carbon builds on Gemini 4 Argon, which the post says launched two weeks earlier. This review looks at what Google reports about Carbon's capabilities, who can use it, and which claims rest only on Google's own testing.

Feature Overview

Gated access through Fairwind. Google says it is "actively engaged in the U.S. government's voluntary process for pre-release model testing" and is still gathering feedback and iterating on guardrails before wider release. Google's Fairwind Program post, published September 2, 2026, describes the program as limited access for governments and trusted partners, with more than 650 participating partners at that time. Participants agree to operating standards such as limiting access to staff in cybersecurity, incident response or penetration testing teams and using multi-factor authentication. The Carbon post does not restate those terms, so whether they apply to Carbon in the same form is not confirmed.

Pricing. Google says Carbon "will launch at the same introductory price as Argon": $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input price. A footnote says $4 per million input and $20 per million output apply after the introductory period; the post does not say how long that period lasts. Because most developers cannot yet call the model, this is a planned price rather than one most readers can currently use.

Reported benchmarks (figures from Google's post).

Google states the external cybersecurity benchmarks were "evaluated without guardrails," meaning they describe the model's raw capability rather than the behavior of the guarded product defenders will use. The Verge reports, from charts in Google's announcement, that Carbon sits ahead of Claude Opus 5.5 and GPT-6 Astra on the Artificial Analysis index and ahead of GPT-6 Astra on SEC-bench Pro.

Cyber defense lifecycle. Google frames the results as covering vulnerability detection, validation and remediation. On internal benchmarks, it reports more than 2x uplift in secure-code generation success over Argon. It also says the Chrome team reported Carbon performed "more than 10X better than another lab's large frontier model" when generating high-quality patches; this is the Chrome team's report as relayed by Google, and the comparison model is not named. Google adds that Wiz observes Carbon leading on web penetration testing benchmarks.

Long-horizon work and alignment. Google says internal teams saw Carbon run for days on multi-step engineering tasks such as root-causing and patching subtle bugs across large codebases and low-level systems. On behavior, it reports successful cheating on Terminal-Bench 4.0 reduced to 0.0% (attempts at 0.4%), high-severity misalignment alerts down almost 80% in live coding environments with internal teams, unsupported claims cut about 2x versus Argon, overreach down 26% and under-service down 29%, and a 0.0% attack success rate on Gray Swan's Indirect Prompt Injection benchmark versus 0.7% for Argon. All of these are Google-reported, and the behavioral figures come from internal testing.

Usability Analysis

For nearly everyone, there is nothing to try yet. The people who can use Carbon today are the subset of cyber defenders Google selected, and the rest of the audience is waiting on guardrail work with no published date.

For those defenders, Google's pitch is a model that spans the whole loop: find a flaw, check whether it is exploitable, and write a fix. The exploitation step is the sensitive one, since a model that scores well on ExploitGym is also capable of the offensive half of the work. That helps explain Google's staged approach, though the post does not itself discuss that dual-use tension in detail. Teams evaluating Carbon once it opens up should plan to test patch quality on their own codebases, since the headline figures come from public benchmarks and Google's internal evaluations.

Pros and Cons

On the plus side, the post reports leading scores across the find, validate and patch stages of defense. The planned price is the same as Argon's. The safety section reports measurable changes in cheating, overreach and prompt-injection resistance, and Google is testing in the U.S. government's pre-release process.

On the limiting side, access is restricted and no general availability date is given. Most evidence is self-reported by Google, with the Chrome team's 10X figure resting on an unnamed comparison model. The benchmark numbers were produced without guardrails, so shipped behavior may differ. The $2/$10 price is introductory, with $4/$20 later.

Outlook

Google says it is eager to see how the initial cohort of cyber defenders uses the model before bringing it to developers, businesses and consumers. Two things to watch are whether independent evaluations reproduce the SEC-bench Pro, ExploitGym and CWE-bench figures, and how much of the capability survives once guardrails are applied. Google did not state when the introductory price window ends.

Conclusion

Gemini 4 Carbon is a cybersecurity-focused frontier model with strong self-reported results, announced alongside a deliberately narrow rollout. It is best read as a preview for security teams, Fairwind partners and government stakeholders, not a model developers can adopt this week. Rating: 4 out of 5, held back by limited availability and the absence of independent verification.

Editor's Verdict

Gemini 4 Carbon: Cyber-Defense Model, Limited Rollout earns a solid recommendation within the Gemini space.

The strongest case for paying attention: reported leading results across finding, validating and patching vulnerabilities. That alone raises the bar for what readers should expect in this space. Reinforcing that, planned introductory pricing matches Argon at $2 input and $10 output per million tokens — practical value rather than just headline appeal. The broader signal worth registering is straightforward: Google's own text describes a gated release, so Carbon should not be treated as a public or API launch despite headlines that say it launched. On the other side of the ledger, one constraint is real rather than a marketing footnote: access is limited to selected cyber defenders and no date is given for developers, businesses or consumers. It should factor into any serious decision. Layered on top of that, most performance evidence is self-reported by Google, and the Chrome team's 10X claim uses an unnamed comparison model — which narrows the set of teams for whom this is an obvious yes.

The cyber defenders in Google's first access group, and Fairwind partners next in line, are the audience that can act on Carbon now; developers and general readers should wait for broader availability and independent evaluations before drawing conclusions.