Google releases Gemini 4 Carbon

Google has resumed its rapid release cadence for frontier AI models. Just two weeks after launching Gemini 4 Argon, the company announced Gemini 4 Carbon. This release is not a lighter Flash variant, but a direct successor to Argon that currently sits behind a staged rollout.

The company claims Carbon represents a major step forward in cybersecurity and agentic tasks. Google states the model shows a significant leap in vulnerability detection, validation, and remediation. According to internal data, Carbon is twice as likely as Argon to produce code free of vulnerabilities. The Chrome team asserts that Carbon generates high-quality patches ten times better than non-Google models.

Performance metrics place Carbon at the top of several internal charts. The model’s score on WebDev Arena increased by 89 points, moving it to the fourth position on the leaderboard. In head-to-head comparisons, Carbon beats Argon by narrow margins in every test. It also outperforms GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1 in nearly all scenarios.

Despite the top rankings, the gap between Carbon and its competitors is small. All frontier models score in a similar range on these specific benchmarks. Google has not clarified whether Carbon is a distinct new model or a post-training update of Argon, noting that the two share many characteristics.

Users cannot access the model immediately. Google is handling the release in stages, a process that may take considerable time. This delayed availability mirrors the approach taken with the earlier Argon release, leaving developers waiting for full public access.