Gemini 4 Carbon Takes Top Spot on AI Index for Cyber Defense
Google unveiled its Gemini 4 Carbon frontier artificial intelligence model on October 29, 2026, targeting complex cybersecurity defense and multi-step software engineering across enterprise systems. Enterprise software developers and defensive cybersecurity teams face immediate changes to vulnerability patching workflows as the model rolls out to select security personnel this week through Google’s Fairwind Program before a broader enterprise release.
New Benchmark Leader Specializes in Vulnerability Detection and Code Repair
Pricing for the new model starts at $2 per million input tokens and $10 per million output tokens, matching the rate set by Gemini 4 Argon two weeks prior, with cached input tokens discounted by 95 per cent. Built by Google DeepMind to work through low-level technical problems across large codebases over continuous days, the system captured first place on the Artificial Analysis Intelligence Index with a score of 59. Standardized evaluation testing recorded an 88.2 per cent score on SEC-bench Pro for identifying software flaws, 54.8 per cent on ExploitGym for testing vulnerability exploitation, and 71 per cent on CWE-bench for automated fix generation. Additionally, attempts to cheat on Terminal-Bench 4.0 dropped to 0.4 per cent with zero successful breaches, alongside a zero per cent attack success rate on the Gray Swan indirect prompt-injection benchmark.
The release follows the deployment of Gemini 4 Argon, continuing a strategy of rolling out targeted frontier systems on accelerated schedules.
