Google announces Gemini 4 Carbon with security gains

Google announced Gemini 4 Carbon on Tuesday, a new frontier AI model that improves on its recent release, Gemini 4 Argon.

The company stated that Carbon shows a significant leap in cybersecurity tasks, specifically in vulnerability detection, validation, and remediation. Google reported that Carbon is twice as likely to generate code free of vulnerabilities compared to Argon. The Chrome team added that the new model is ten times better at generating high-quality patches than non-Google alternatives. Carbon also raised its score by 89 points on WebDev Arena, moving it to the fourth position on the leaderboard.

This announcement came just two weeks after Google released Gemini 4 Argon. While the two models share many similarities, Carbon surpasses Argon by slim margins in every test Google provided. The new model also outperforms GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1 in almost every benchmark, though the gaps between these top-tier models remain small.

Google is treating this as a distinct new model rather than a simple update, yet the performance differences suggest it may be a post-training upgrade of the previous version. The company is implementing a staged release process, meaning users cannot yet access the tool. This approach mirrors the rollout strategy used for Argon and could take considerable time to complete.

The announcement follows a period where Google paused releases of new frontier models earlier this year. The company now positions Carbon as a tool that reaches new thresholds in coding, agentic tasks, and scientific analysis, even though the public must wait for the staged rollout to finish.