Google introduces Gemini 4 Carbon for cybersecurity and long-horizon work

Google has announced Gemini 4 Carbon, a new frontier model focused on cybersecurity, coding and long-horizon agentic work. The announcement was published on 28 October 2026 by Google, which describes Carbon as the successor to Gemini 4 Argon, released two weeks earlier. The model is not generally available yet: Google says it is starting with a subset of cyber defenders and will expand access through its Fairwind Program before bringing Carbon to developers, businesses and consumers.

Google reports that Carbon ranks first on the recently updated v5.0 Artificial Analysis Intelligence Index, with a score of 59. It also reports results of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for validating whether vulnerabilities can be exploited, and 71.0% on CWE-bench for producing patches. These are reported benchmark results from Google’s announcement, not an independent assessment of every real-world security scenario.

The model is designed for sustained, multi-step work. Google says internal teams have seen it continue for days through dead ends while root-causing and patching subtle bugs in large codebases and low-level systems. In web development, the company reports an 89-point increase on WebDev Arena, placing Carbon fourth at 1768 at the time of publication. Google also describes internal improvements in secure-code generation, unsupported claims, overreach and resistance to indirect prompt injection. It reports a zero percent attack success rate on the evaluated Gray Swan benchmark, but that result should not be treated as proof of universal prompt-injection protection.

For AI users and makers, the important change is the combination of longer-running workflows with a deliberate, limited rollout. Cyber defenders may get earlier access to a model that can help locate, validate and repair vulnerabilities, while developers and companies still have to wait for broader availability. Google says Carbon will use the same introductory pricing as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95% once it launches. The announcement therefore signals competition around reliable software agents, but the practical value will depend on independent testing, access terms and safeguards outside Google’s internal evaluations.