Google unveiled Gemini 4 Carbon on Oct. 28 as a model for cybersecurity defense and extended coding work, with access initially limited to selected cyber defenders.
The launch came about two weeks after Gemini 4 Argon. It gives Google another entry in the competition among technology companies to build advanced AI tools for software engineering and security. Google says Carbon can help find software vulnerabilities, test whether they can be exploited and develop fixes.
The company reported scores of 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench. These are benchmark results reported by Google. Google said the evaluations were done without safeguards.
That detail matters when judging how the model might perform in real-world use. Google also says Carbon generated secure code at more than twice Argon’s success rate in internal tests. Google teams have used the model for extended engineering tasks.
These tasks included tracing subtle bugs and fixing them across large codebases. The company says Carbon can keep working through problems for days. Access is starting with a subset of cyber defenders.
Google expects to expand access to more trusted partners through its Fairwind Program before making the model available to developers, businesses and consumers through Google products. The company says it is improving safeguards. Google is also taking part in a U.S. government voluntary process for testing models before release.
Introductory API pricing is $2 per million input tokens and $10 per million output tokens.