Google announces Gemini 4 Carbon AI model with cybersecurity focus

Google has announced the release of Gemini 4 Carbon, a new frontier AI model that provides improvements in coding, agentic tasks, and scientific analysis [4, 46]. This model is an upgrade to Gemini 4 Argon, which was released by the company two weeks ago [4, 39, 46].

A major focus of the new model is its performance in cybersecurity [4, 46]. According to Ars Technica, Google claims the model provides a major advancement in the detection, validation, and remediation of vulnerabilities [46]. The company reports that Carbon is twice as likely to generate code that is free of vulnerabilities compared to the Argon model [46]. Additionally, Google's Chrome team asserts that Carbon is 10 times more effective at generating high-quality patches than models from non-Google developers [46].

The model's performance has been highlighted through various industry benchmarks [4, 46]. Google states that Carbon has secured the top position on the Artificial Analysis Intelligence Index (v5.0), which places it ahead of other models such as Claude Opus 5.5 and GPT-6 Astra [4]. Additionally, the model has shown progress on the WebDev Arena, where it reportedly increased its score by 89 points to reach the fourth position on the leaderboard [46].

Gemini 4 Carbon is not yet available for general use [4, 39, 46]. Currently, the model is only accessible to a select group of cybersecurity defense companies [4]. Google intends to eventually expand access to other partners through its Fairwind Program [4].