Google introduces Gemini 4 Carbon — flagship focused on cybersecurity

Google introduced the flagship Gemini 4 Carbon model, released two weeks after Gemini 4 Argon and ranking 1st in the third-party Artificial Analysis Intelligence Index v5.0 with a score of 59, with a primary focus on cybersecurity and phased access availability at an introductory API price of $2/$10 per million input and output tokens.

Compiled by Sergey KostenchukPublished 2026-10-28Updated 2026-10-28

2026-10-28 Research Google

🤖 Google introduces Gemini 4 Carbon

Released two weeks after Gemini 4 Argon and ranked 1st in the third-party Artificial Analysis Intelligence Index v5.0 (59 points). The focus is on cybersecurity: 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71% on CWE-bench — finding, verifying, and fixing vulnerabilities. Access is phased: first security specialists, then Fairwind partners, with a general release later. Initial API price is $2/$10 per million tokens.

🌍 For the first time, a Google flagship is positioned as a defense tool: a full cycle of 'find → verify → fix'; 0.0% attack success rate in Gray Swan's indirect prompt injection test compared to 0.7% for Argon.

👤 Focus on agents working autonomously for days: they independently find the cause of a bug and write a patch; according to Google's tests, safe code is more than twice as good as Argon's, with approximately 80% fewer serious instruction violations.

Source 1: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/

Sources