Google introduces Gemini 4 Carbon with a focus on cybersecurity

Google has unveiled Gemini 4 Carbon, a new frontier AI model designed for complex software engineering and cybersecurity work. Koray Kavukcuoglu announces the model in an official Google blog post, describing it as the company’s strongest model so far for cybersecurity and long-running tasks.

The company says Carbon ranks first on version 5.0 of Artificial Analysis’ Intelligence Index, with a score of 59. Google also reports strong results on security tests that measure vulnerability discovery, exploit validation, and code patching. These figures include 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. Google notes that these external benchmark results were measured without the model’s safety guardrails.

Built for extended engineering tasks

Carbon is intended to work through multi-step projects over long periods. Google says its internal teams have used the model for days at a time to investigate subtle bugs, trace their causes, and prepare fixes across large codebases and low-level systems.

The company also reports improvements in web development, scientific tasks, multimodal understanding, and agentic workflows. On WebDev Arena, Google says Carbon reached fourth place with a score of 1,768, an increase of 89 Elo points from its earlier model, Gemini 4 Argon.

Safety claims accompany stronger cyber capabilities

Google says it has strengthened alignment and misuse protections alongside the model’s cyber capabilities. According to the company, Carbon reduced unsupported claims by about half compared with Argon and lowered high-severity misalignment alerts by nearly 80% in internal live coding tests.

Google further says Carbon resisted all attacks evaluated in Gray Swan’s Indirect Prompt Injection benchmark. The model will initially be available to a limited group of cyber defenders and trusted partners through Google’s Fairwind Program. Google plans broader access later.

During the introductory period, Carbon costs $2 per million input tokens and $10 per million output tokens. Google says the prices will later rise to $4 and $20 respectively.