ModelGemini

Google launches Gemini 4 Carbon, its strongest model

Original title: Gemini 4 Carbon: Our next frontier model

Plain language

What it is: Google announced Gemini 4 Carbon, its strongest AI model designed for cybersecurity and complex, long-term tasks.
How it works: The model can run for days to solve multi-step engineering problems, find security vulnerabilities, and write secure code. It is also trained to be more honest and reduce errors or fabricated information.
Where it is available: It is currently available to a select group of cybersecurity defenders and trusted partners through the Fairwind Program, with plans to release it to developers and businesses later.
Example: The Chrome team used the model to generate high-quality software patches to fix security bugs.

What it is: Google DeepMind announced Gemini 4 Carbon, a frontier model featuring advanced cybersecurity capabilities and long-horizon task execution.
How it works: It uses a multi-layered safety stack, resists prompt injections, and is optimized for agentic workflows to patch bugs across large codebases. It reduces cheating on benchmarks and lowers high-severity misalignment alerts in live coding.
Where it is available: It is initially rolling out to select cyber defenders and partners in the Fairwind Program, and will later be available via APIs and Google product surfaces.
Example: The Chrome team utilized Carbon to generate high-quality patches, achieving a performance over ten times better than a competitor's large model.

The headline and explanation on this page are based on Google's post. The full details are in the original.

Gemini 4 Carbon: Our next frontier model