Google has unveiled Gemini 4 Carbon, its latest frontier artificial intelligence model, introducing major advances in cybersecurity, software engineering and complex autonomous computing tasks.
Announced on October 28, 2026, the model builds on Gemini 4 Argon, released two weeks earlier, as Google accelerates the development of advanced AI systems for enterprise and technical applications.
According to Google, Gemini 4 Carbon delivers improved capabilities in identifying software vulnerabilities, assessing security risks and generating patches for complex coding issues.
The company reported an 88.2% score on SEC-bench Pro for vulnerability detection, alongside 54.8% on ExploitGym and 71.0% on CWE-bench, demonstrating progress across several cybersecurity benchmarks.
Beyond security applications, the model is designed to handle extended software engineering tasks, including investigating technical problems, managing complex workflows and implementing solutions across large codebases.
Google also reported improvements in secure code generation compared with Gemini 4 Argon, alongside reduced error rates in internal coding environments.
Initial access will be restricted to selected cybersecurity defenders, with additional trusted partners gaining access through Google's Fairwind Program. Broader availability for developers, businesses and consumers is planned following further safety evaluations.
The controlled rollout reflects the growing importance of safeguards as advanced AI systems become increasingly capable of performing sophisticated cybersecurity tasks.
The announcement strengthens Google's position in the competitive AI market, highlighting its focus on combining advanced reasoning, software development and cybersecurity capabilities within its next generation of AI models.
