01The key point
Google's most capable model yet is designed for cybersecurity and long-running tasks, but most people cannot use it yet.
02What happened
Google announced Gemini 4 Carbon (Gemini 4 Carbon: our next frontier model). The company says it was designed for long-running, multi-step engineering work. Its internal teams have seen it run continuously for several days, finding the root causes of bugs and fixing them in large codebases. Google reported a score of 59, ranking first on the Artificial Analysis Intelligence Index v5.0, and 88.2% on SEC-bench Pro, a vulnerability-finding benchmark.
The blog says Carbon will first be available this week to some cyber defenders, with access then widening through Fairwind to additional trusted partners. It also says developers, businesses, and general users should get access as soon as possible, but gives no date. Promotional pricing is $2 per million input tokens and $10 per million output tokens; after the promotion, prices will rise to $4 and $20, respectively.
03What this means for you
Most people cannot use it yet. If you want to try it, watch for official availability announcements. If you later use it for long-running coding agents, remember that prices will double after the promotional period. All scores were reported by Google (🗣 Self-reported); the cybersecurity tests did not have safeguards enabled, and this site has not tested the model.
04Correction (2026-10-30)
The original article said trusted cybersecurity defenders in Fairwind would get access first and attributed that claim to Google AI’s official account. The blog instead describes an initial release to some cyber defenders this week, followed by broader access for trusted partners through Fairwind. The article has been corrected.
05Sources
- Gemini 4 Carbon: our next frontier model: Google official blog, published 2026-10-28 at 17:00 UTC (01:00 Taipei time); announces Gemini 4 Carbon, its scores, rollout order, and pricing