Gemini 4 Carbon is a frontier AI model from Google DeepMind that Google calls its strongest, built for long, many-step tasks and for finding and fixing security flaws.
In plain words
Carbon follows Gemini 4 Argon, which Google released two weeks earlier. It is built for agentic work, where a model plans and acts over many steps, and Google says its teams have seen it keep going for days on hard engineering problems. It covers each stage of cyber defence: spotting a weakness, checking whether it can be exploited, and fixing it. Cyber defenders got it first through Google's Fairwind Program; two weeks later it opened to paying developers via the Gemini API and to Google AI Ultra subscribers.
A quick example
Imagine a security team pointing Carbon at a large codebase. It hunts for a hidden flaw, checks whether an attacker could really use it, then writes a patch that fixes the cause.
Where you'll meet it
- Long, multi-step coding and engineering work.
- Developer apps built on the Gemini API.
- Cyber defence through the Fairwind Program.
Related
Sources
- DocsGemini 4 Carbon: Our next frontier model, Google DeepMind · read 28 Oct 2026
- DocsGemini 4 Carbon: our next frontier model, Google · read 28 Oct 2026
- DocsGemini 4 Carbon is now available, Google · read 10 Nov 2026
Checked by WhatIsAI against the sources above · 10 Nov 2026. Spotted an error? Report it.