Google introduces Gemini 4 Carbon: top model for cybersecurity

Google has presented its latest AI model, Gemini 4 Carbon, which performs particularly well in the field of cybersecurity and long-term engineering tasks. The announcement took place on October 28, 2026, following the recent launch of Gemini 4 Argon just two weeks earlier. Initially, Carbon will be made available to selected cybersecurity experts through the Fairwind program, with future availability in the Netherlands and support for the Dutch language not yet confirmed.

With a score of 88.2% on the SEC-Bench Pro for finding vulnerabilities and a 54.8% score on ExploitGym, Carbon demonstrates impressive capabilities in identifying and exploiting vulnerabilities in software. Google describes the model as being safer than its predecessor, with significantly fewer unfounded claims and unauthorized actions. This makes it a valuable tool for software developers and security teams working on complex tasks.

Prices for access to Gemini 4 Carbon via the API start with an introductory rate of €1.80 per million input tokens and €8.98 per million output tokens. After the introductory period, these prices will increase. Access to the system will be phased, and it remains unclear to what extent and when the model will be available to a wider audience, including developers and companies.

Currently, there is uncertainty regarding the specific availability of Gemini 4 Carbon in the Netherlands and in Dutch. For now, only selected cybersecurity experts will have access to the model, and it is unclear when other users, including companies and consumers, will be able to leverage this technology.

Read the full article from AI Insider.