Google has introduced Gemini 4 Carbon, its newest frontier model and the second Gemini 4 release in a fortnight, following Gemini 4 Argon. The company describes Carbon as its strongest model to date, built for long, demanding engineering work and for cybersecurity defense.
The headline numbers come from security testing. Google says Carbon scored 88.2% on SEC-bench Pro, which measures finding real vulnerabilities in software, 54.8% on ExploitGym, which tests whether flaws can actually be exploited, and 71.0% on CWE-bench, which measures turning confirmed flaws into working fixes. It also ranks first on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59.
For everyday engineering, Google says Carbon is designed to stay on hard, multi-step problems for days, working through dead ends to trace and patch subtle bugs across large codebases. It also reports improvements in web development quality, with Carbon reaching fourth place on the WebDev Arena leaderboard.
Access is deliberately limited for now. Starting this week, Google is offering Carbon to a subset of cyber defenders, then gradually expanding to additional trusted partners through its Fairwind program, while it takes part in the U.S. government’s voluntary pre-release testing process. Developers, businesses, and ordinary users will have to wait for a wider release.
The pricing, when it arrives, is set to match Argon at $2 per million input tokens and $10 per million output tokens. For teams that already lean on AI-assisted coding and security review, tools like Gemini, Gemini Code Assist, Snyk, and Semgrep cover much of the same ground today, from code generation to vulnerability scanning.
Source: Google Blog
