Google announced Gemini 4 Carbon on October 28, describing it as its strongest model for cybersecurity and long-running tasks. Access is starting with a selected group of cyber defenders through Google’s Fairwind program. Developers, businesses and consumers do not have general access yet; Google says it plans to expand access after gathering feedback and improving safeguards.
The model also ranks first on Artificial Analysis’ Intelligence Index V5, according to Google. Tom’s Hardware reports that Carbon scored 59, one point ahead of Claude Opus 5.5, and placed first in four of the index’s nine evaluations. It also topped Artificial Analysis’ AA-Omniscience v1.1 test, which measures hallucinations and recall of known information. That result is a benchmark ranking, not a guarantee that Carbon will avoid mistakes in everyday use.
What do the security tests show?
Google reports scores of 88.2% on SEC-bench Pro, for finding software vulnerabilities; 54.8% on ExploitGym, for testing whether flaws can be exploited; and 71.0% on CWE-bench, for fixing vulnerabilities. Google says these tests were run without its guardrails, and calls the results state of the art. The company also says internal tests found Carbon more than twice as successful as Gemini 4 Argon at generating secure code. Google’s Chrome team says Carbon performed more than 10 times better than another company’s frontier model at producing high-quality patches, but it does not identify that model or provide further test details in the announcement.
Google says Carbon improved by 89 points on WebDev Arena, reaching fourth place with a score of 1,768. It also describes the model as designed for complex, multi-step work that may continue for days. These are company-reported results and capabilities; the announcement does not establish how the model will perform across all users’ coding or security tasks.
When will it reach more users?
Google says Carbon will be offered to additional trusted partners gradually, then made available to developers, businesses and consumers through its products “as soon as possible.” It has not given a date for that broader launch. Ars Technica notes that Carbon follows Gemini 4 Argon by about two weeks, while Google describes it as building on Argon.
Google says the introductory price will match Argon: $2 per million input tokens and $10 per million output tokens. Cached input tokens will cost 95% less than the standard input rate. Those prices are useful for developers planning future use, but they do not mean the model is currently open to them.
Why it matters
Cyber defenders in Google’s initial Fairwind group are the first people who can try Carbon; developers and ordinary users still have no general access date. Google has announced prices for a later launch, but whether its benchmark gains translate into better results for a particular user’s work remains untested here.
