Google announced Gemini 4 Carbon, a new frontier AI model that improves on Gemini 4 Argon released two weeks earlier. The model is not yet available to the public.
Carbon shows gains in coding, agentic tasks, and scientific analysis. Google claims the model makes a "significant leap" in vulnerability detection, validation, and remediation for cybersecurity applications. Carbon is twice as likely to generate code free of vulnerabilities compared to Argon, and the Chrome team says it is 10 times better at generating "high-quality" patches versus non-Google models. Carbon scored 89 points higher on WebDev Arena and now ranks fourth on that leaderboard.
On Google's benchmarks, Carbon beats GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1 in nearly every test, though the gaps between models are small. In the Grey Swan evaluation measuring prompt-injection attack resistance, Carbon scored 0 percent, better than Argon at 0.7 percent, Opus 5.5 at 1 percent, and GPT-6 Astra at 8.5 percent.
Google is taking a "deliberative approach" to Carbon's release, working within a voluntary testing framework established by the U.S. government. The company plans to roll out the model through its Fairwinds cyber defense program before wider availability. Carbon will likely reach AI Ultra subscribers and Gemini API developers first. Google is offering an introductory API rate of $2 per million input tokens and $10 per million output tokens, which will double after the promotional period ends. The company has declined to specify when that increase will occur.