Gemini 4 Carbon (Limited Preview)
Google's new top model, built for long multi-step work and cybersecurity, currently open only to a small group of vetted security teams.

Details
- Google says internal teams watched Carbon work for days on a single engineering task, working through dead ends without being restarted
- It follows Gemini 4 Argon, which launched two weeks earlier, and uses the same safety setup plus fewer unsupported claims and better resistance to hidden instructions planted in content it reads
- Access starts with a subset of cyber defenders and expands to other trusted partners through Google's Fairwind Program. Developers, businesses, and consumers get it later, with no date given
- Introductory API pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Cached input costs 95% less
Best for
Long, multi-step technical jobs like finding and fixing bugs across a large codebase, where the model has to keep going through dead ends.
When to use it
Watch for it now and plan for it later. Most teams can't use it yet, but it signals what Google's next generally available model will be able to do.
Who benefits
Security teams first, then developers and businesses on Google's products once access widens.
Join the newsletter to get ahead with AI.
I share strategies, new modules, free guides, and tools to get you better at using AI in your work.
Leading a team? See team programs →

