Google Withholds New AI Model From Public
Gemini 4 Carbon tops coding benchmarks, yet access remains restricted pending a staged rollout.
By Lauren Bello ·

Google announced Gemini 4 Carbon on October 28, 2026. The release follows Gemini 4 Argon, which launched just two weeks prior. This new model is not the Flash variant users anticipated. Instead, it represents an improvement over Argon. Google plans a staged release process that could extend over a long period.
The company claims Carbon marks a new threshold in coding, agentic tasks, and scientific analysis. Google stated the model achieves a significant leap in vulnerability detection, validation, and remediation for cybersecurity applications. Data indicates Carbon is twice as likely to generate code free of vulnerabilities compared to Argon. The Chrome team reported that Carbon generates high-quality patches at a rate ten times higher than non-Google models.
Performance metrics show Carbon ranked fourth on the WebDev Arena leaderboard. The model increased its score by 89 points in that specific arena. Google presents Carbon as a distinct, new model. However, it may function as a post-training upgrade to Argon. The two models share many similarities, though Carbon outperforms Argon by small margins in every test provided.
In comparisons against competitors, Carbon beats GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1 in nearly all tests. The lead over these rivals consists of only a few points. All frontier models currently score within the same general range on these benchmarks. Google has not set a specific date for public access to the model.





