Gemini 4 Carbon: Google’s Top-Ranked AI Model Goes to Cyber Defenders First

Google’s Gemini 4 Carbon scores 59 on the Artificial Analysis Intelligence Index, the top score, but only vetted cyber defenders can use it. Its earlier model Argon reached developers 13 days after defenders.
artificial-intelligence
cyber-security
Author

Kabui, Charles

Published

2026-11-02

Keywords

gemini-4-carbon, google-deepmind, fairwind-program, cyber-defenders, ai-cybersecurity, staged-release

Download as PDF

Gemini 4 Carbon is Google’s strongest AI model, and outside Google only vetted cyber defenders can use it. These are security teams at governments, hospitals and phone networks in Google’s Fairwind Program, which aims to give 650-plus partners a head start on attackers. Google announced Carbon on October 28 with no public API or release date. Independent tester Artificial Analysis ranks it first at 59 on its Intelligence Index, which combines nine tests of coding, knowledge and reasoning. Claude Opus 5.5 scores 58 and Google’s earlier Argon 50.

Google’s security scores, each the share of tasks solved, come from runs without safety limits. On SEC-bench Pro, where an AI hunts real flaws in browser and Linux code, Carbon scored 88.2%. ExploitGym asks it to turn a known flaw into a working attack, and it scored 54.8%. On CWE-bench, which asks it to fix flaws in real code, it scored 71.0%. On Gray Swan’s test of instructions hidden in emails and web pages, Google says 0.0% of attacks worked.

Google gave Argon to defenders on September 30 and to developers 13 days later, with cyber safeguards on. Carbon arrived 15 days after Argon went public and is still defenders only. CEO Sundar Pichai says wider availability is coming soon. OpenAI’s Daybreak and Anthropic’s Glasswing give trusted companies early access. Google’s launch price is $2 per million input tokens (text chunks) and $10 per million output, half the $4 and $20 that Artificial Analysis lists.

Read More: Gemini 4 Argon, the model Google gave to defenders first on September 30, Gemini 3.8 Flash Cyber, Google’s earlier defender-only model, Anthropic’s free bug scanner for open-source projects and Anthropic’s unreleased model that links bugs into attacks on five major systems

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {Gemini 4 {Carbon:} {Google’s} {Top-Ranked} {AI} {Model}
    {Goes} to {Cyber} {Defenders} {First}},
  date = {2026-11-02},
  url = {https://toknow.ai/posts/gemini-4-carbon-cyber-defenders-first-59-index/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “Gemini 4 Carbon: Google’s Top-Ranked AI Model Goes to Cyber Defenders First.” https://toknow.ai/posts/gemini-4-carbon-cyber-defenders-first-59-index/.