Following the recent rollout of Gemini 4 Argon, Google has announced Gemini 4 Carbon as its latest frontier artificial intelligence model. Much like its predecessor, the new model features a staged release schedule that will take some time to fully deploy, meaning users cannot access it immediately.
Improvements in Coding and Cybersecurity
Google states that its newest model achieves new milestones in coding capabilities, agentic tasks, and scientific analysis. Specifically, the system delivers substantial enhancements in cybersecurity applications, demonstrating a notable advancement in vulnerability detection, validation, and remediation.
According to Google, Carbon is twice as likely to generate code free of vulnerabilities compared to Argon. Additionally, the Chrome team reports that Carbon performs 10 times better at producing high-quality patches than non-Google models. The system also increased its score by 89 points on WebDev Arena, securing the fourth position on the leaderboard.
Performance Benchmarks and Competition
Google presents Carbon as an entirely new model, though it may function as a post-training upgrade to Argon given their shared traits. In company-provided testing, Carbon outperforms Argon by narrow margins across all evaluations.
The new model also surpasses GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1 in nearly every test. However, all major frontier models score within a similar range on these benchmarks, with the performance gap between Carbon and competing systems spanning only a few points.
