From the source
Google has announced Gemini 4 Carbon, a frontier model it says improves on Gemini 4 Argon in coding, agentic tasks and scientific analysis, but which it is not yet letting the public use.
- Google says Carbon shows a "significant leap" in vulnerability detection, validation and remediation, and is twice as likely as Argon to generate code free of vulnerabilities.
- The Chrome team claims Carbon is 10 times better at generating "high-quality" patches than non-Google models, and Google says its WebDev Arena score rose 89 points to rank fourth.
- Google reports Carbon resisted every attempted prompt-injection attack in the Grey Swan evaluation, a 0 percent score against 0.7 percent for Argon, 1 percent for Opus 5.5 and 8.5 percent for GPT-6 Astra.
- The company says the model's agentic coding and cybersecurity performance requires a "deliberative approach", and it is working within the voluntary testing framework established by the US government before rolling Carbon out through its Fairwinds cyber defense program.
- Google gives no timeline for a wide release, but says Carbon will probably arrive first for AI Ultra subscribers and Gemini API developers, with a temporary discounted rate of $2 per million input tokens and $10 per million output tokens that will double after the introductory period.