Google Unveils Gemini 4 Carbon AI Model for Enhanced Cybersecurity and Code Efficiency

The Gemini 4 Carbon AI model, developed from the previous Argon model, focuses on improving cybersecurity and code efficiency. It has been tested on various benchmarks, including the Artificial Analysis Intelligence Index (v5.0), where it outperformed Claude Opus 5.5 and GPT-6 Astra. The model has been designed to reduce errors in complex engineering tasks and improve cybersecurity capabilities, including vulnerability detection and exploitation prevention.

In cybersecurity assessments, the Gemini 4 Carbon model achieved a top score in the SEC-bench Pro test, which evaluates the ability to detect vulnerabilities and errors in real-world systems. It also performed well in the ExploitGym test, which assesses the ability to analyze vulnerabilities and generate test cases for system exploitation.

The model’s architecture is optimized for complex engineering tasks, reducing errors caused by sequential processing and improving resistance to prompt injection attacks. It also reduces the rate of generating unverified data by two times compared to the Argon model, with a processing cost of $2 per 1 million input tokens and $10 per 1 million output tokens.