Google Announces Gemini 4 Carbon, a Cybersecurity Focused Model Held From Public Access

Google announced Gemini 4 Carbon on Tuesday, a new frontier AI model focused on cybersecurity tasks. The release follows Gemini 4 Argon, which launched just two weeks prior. Google is applying a staged release strategy to Carbon, meaning general public access is not yet available.

The company stated that Carbon represents a major advancement in vulnerability detection, validation, and remediation. According to Google, the model is twice as likely as Argon to generate code free of vulnerabilities. The Chrome team added that Carbon produces high-quality security patches ten times more effectively than non-Google models.

Performance metrics place Carbon at the top of current leaderboards. The model increased its score on WebDev Arena by 89 points, securing the fourth position on the leaderboard. In head-to-head comparisons, Carbon outperformed GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1 in nearly all tested categories.

Despite these rankings, the gap between Carbon and other leading models remains narrow. Google described the improvements as a new threshold in coding and scientific analysis, though the company did not clarify if Carbon is a distinct architecture or a post-training upgrade of Argon. The two models share many similarities, with Carbon holding slim margins in every benchmark provided.

Users must continue to wait for the full rollout. Google confirmed that the model is currently undergoing a phased distribution process. This approach mirrors the launch timeline used for the earlier Argon release, extending the period before the model becomes widely accessible.