Google announced Gemini 4 Carbon on Oct. 28, 2026, describing it as its strongest model and saying it is designed for sustained agentic workflows and complex, long-horizon engineering tasks. The company says it ranked first on the Artificial Analysis Intelligence Index v5.0 with a score of 59 and placed fourth on WebDev Arena at 1,768 after gaining 89 Elo points.
Google reports Carbon scored 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching. It says internal benchmarks showed more than twice the secure-code-generation success rate of Gemini 4 Argon, while testing also found lower unsupported claims and reduced prompt-injection attack success; the company notes the external cybersecurity benchmark results were evaluated without guardrails.
Google says Carbon will first be offered to a subset of cyber defenders, with access expanding to trusted partners through its Fairwind Program while it gathers feedback and iterates on safeguards. It plans to make the model available to developers, businesses and consumers through its products as soon as possible, at an introductory price of $2 per million input tokens and $10 per million output tokens; cached input tokens cost 95% less than the input rate. The page does not specify when broader access will begin.
For AI builders, Carbon’s reported long-running engineering and security capabilities could support more autonomous coding and vulnerability-remediation workflows. A key caveat is that availability is initially restricted, and the cited benchmark results do not establish performance in every real-world deployment.