Google has unveiled a new AI model, 'Gemini 4 Carbon', that boosts coding, agentic tasks and cybersecurity performance.

The Verge reported on Oct. 28 that Gemini 4 Carbon ranked first in the Artificial Analysis Intelligence Index (v5.0), beating Anthropic Claude Opus 5.5 and GPT-6 Astra.

Gemini 4 Carbon is an improved version of Argon, which Google unveiled earlier this month. It is currently available only to some cybersecurity defence companies and will later be expanded to other partners.

Google is placing particular emphasis on the model's security performance. Gemini 4 Carbon improves vulnerability detection and secure code generation, and also ranked first on SEC-Bench Pro, which evaluates bug-finding ability in real-world environments, and ExploitGym, which assesses the ability to turn vulnerabilities into exploits, the company said.

Its ability to carry out long-term tasks has also been strengthened.

A Google internal team said Gemini 4 Carbon continued working for days even when it hit dead ends while solving complex, multi-step engineering tasks. It added that resistance to prompt-injection attacks and the ability to collaborate with users have also improved.

Pricing is at the same level as Argon. It costs $2 per 1 million input tokens and $10 per 1 million output tokens.