🇺🇸 English 🇰🇷 한국어 🇯🇵 日本語
Home › AI & Machine Learning
AI & Machine Learning

Gemini 4 Carbon Goes to Cyber Defenders First

Google unveiled Gemini 4 Carbon on Oct. 28, reporting 88.2% on SEC-bench Pro and a top score of 59 on the Artificial Analysis index, and gave first access to cyber defenders via its Fairwind Program.

Google unveiled Gemini 4 Carbon on Wednesday, calling it its strongest model yet and handing it first to a small group of cyber defenders rather than to developers or consumers. In a blog post signed by Koray Kavukcuoglu, senior vice president of Google DeepMind and Google's chief AI architect, the company said Carbon is built on Gemini 4 Argon and posts state-of-the-art results on vulnerability-hunting benchmarks.

Colorful lines of source code on a dark computer screen
Lines of source code on a monitor. File photo · 사진: Ilya Pavlov ilyapavlov / 위키미디어 공용(CC0 1.0)

Key takeaways

  • Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym and 71.0% on CWE-bench for patching, all measured without guardrails.
  • Access starts this week with a subset of cyber defenders, then widens to vetted partners through Google's Fairwind Program; no date is set for developers or consumers.
  • Introductory pricing matches Argon at $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20.

Gemini 4 Carbon benchmark results

Google's headline claim is a first-place score of 59 on version 5.0 of the Artificial Analysis Intelligence Index, a composite ranking of frontier models. On WebDev Arena, Carbon gained 89 Elo points to reach 1769, which places it fourth rather than first.

The security numbers are where Google is staking its lead. The company said Carbon's cybersecurity scores were evaluated without guardrails.

Gemini 4 Carbon security benchmarks (Google, Oct. 28, 2026)
BenchmarkTaskScore
SEC-bench ProFinding vulnerabilities88.2%
ExploitGymExploitation54.8%
CWE-benchPatching71.0%

Google also cited internal results: more than double Argon's success rate on secure-code generation, and a report from its Chrome team that Carbon was more than 10 times better than "another lab's large frontier model" at producing high-quality patches. The company did not name that rival, and none of these figures has been independently reproduced.

Is Gemini 4 Carbon safer than Argon?

Google says yes on several of its own measures. Carbon's attack success rate on Gray Swan's indirect prompt-injection benchmark fell to 0.0%, from 0.7% for Argon. On Terminal-Bench 4.0, successful cheating dropped to 0.0% and cheating attempts to 0.4%.

In internal live-coding tests, Google said high-severity misalignment alerts fell by almost 80%. It also reported about half as many unsupported claims as Argon, a 26% reduction in overreach and a 29% reduction in what it calls under-service.

Who can use Gemini 4 Carbon

Google is repeating the cyber-first rollout it used for Argon, which reached Fairwind members on Sept. 30. The Fairwind Program prioritizes governments and national cyber authorities, critical infrastructure operators in healthcare, telecom, energy and finance, and core technology platforms. Applicants are vetted, including background checks, and access is limited to internal security, incident response and penetration testing teams.

Google said it is taking part in the U.S. government's voluntary process for pre-release model testing. Broader availability for developers, businesses and consumers will follow "as soon as possible," the company said, after more feedback and work on guardrails.

Why it matters

Carbon arrives as AI labs compete on who can find and fix software flaws fastest, and as they debate who should get those capabilities first. Google's approach keeps the unrestricted model inside a vetted circle of defenders while the general release waits. Kavukcuoglu signaled that approach earlier, when he said Gemini 4 was in refinement and would ship early.

What's next

The key open questions are when paying API customers get access and whether outside evaluators confirm Google's numbers. Cached input tokens will be billed at a 95% discount, and the $4 and $20 standard rates take effect after an introductory period that Google has not dated.

FAQ

Can developers use Gemini 4 Carbon today? No. Google is starting with a subset of cyber defenders and will phase in trusted partners through Fairwind. It has not given a date for general developer access.

How much will Gemini 4 Carbon cost? Introductory pricing is $2 per million input tokens and $10 per million output tokens, the same as Argon's launch price. It rises to $4 and $20 after the introductory period.

How do organizations apply for Fairwind? Google directs applicants to their Google Cloud account team. Organizations that are not yet Google Cloud customers start with Google Cloud's onboarding process.

References

#Gemini 4 Carbon #Google DeepMind #AI cybersecurity #Fairwind Program #Frontier Model #AI Benchmarks
S
Seung Jung
AI Newsway · Developer

Editor responsible for AI Newsway. Writes about AI, developer tools, cloud and productivity technology.

More by this reporter ›
Copyright © AI Newsway All rights reserved. Unauthorized reproduction, redistribution, and AI training prohibited.
Share
f X K

Comments 0

Comments are currently disabled.

Be the first to comment.