Google Releases Gemini 4 Carbon with Enhanced Network Security Capabilities

Google has released its latest model, Gemini 4 Carbon, which has made significant improvements in code writing, AI tasks, and network security. The model has topped the Artificial Analysis Intelligence Index (v5.0) benchmark list, with Claude Opus 5.5 and GPT-6 Astra following closely. This new model has upgraded performance compared to the Argon model released in October.

Gemini 4 Carbon has shown significant advancements in network defense capabilities, particularly in vulnerability discovery and secure code generation. It has topped two benchmark tests: SEC-bench Pro, which evaluates AI models’ ability to discover real vulnerabilities, and ExploitGym, which measures the model’s ability to convert vulnerabilities into exploit programs.

Google claims that Gemini 4 Carbon is designed for long-cycle tasks and has been observed to run continuously for several days, even when encountering dead ends, and handle complex multi-step engineering tasks. It is also more resistant to shortcut attacks and prompt injection attacks, and has stronger human-machine collaboration capabilities compared to Argon.