NEWS / OCT.2026
Google announces Gemini 4 Carbon with initially restricted access
Google announced Gemini 4 Carbon, an artificial intelligence model focused on cyberdefense, on October 28, 2026. It will initially be available to a select group of specialists, before access is gradually expanded.

Google announced Gemini 4 Carbon with limited initial access
Google announced Gemini 4 Carbon on October 28, 2026. The company plans to give a select group of cyberdefense specialists access to the model during the week of the announcement.
Google then plans to expand access to trusted partners through Fairwind, followed by developers, businesses, and consumers, with no firm date for the wider rollout. The company presents the initial phase as an opportunity to gather feedback and work on the model’s safeguards.
Carbon is based on Gemini 4 Argon, which is already used for security tasks. Google announced Argon on September 30, then made it more widely available on October 13, about two weeks before the announcement of Carbon.
For security teams, Google describes a vulnerability handling workflow. After identifying a flaw, Carbon can help verify whether it is exploitable, so that risks can be prioritized, and then suggest a fix. This sequence could speed up defenders’ work, although the published figures do not establish how much time it saves in real-world conditions.
Google claims scores of 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for remediation. The company says these evaluations were conducted without safeguards. The scores therefore do not guarantee equivalent performance with protections enabled in production.
Meanwhile, Artificial Analysis gives Carbon 59 points in high reasoning mode on its Intelligence Index v5. This independent index evaluates the model generally; it does not, on its own, measure its performance in operational cyberdefense. Independent evaluations involving real incidents and fixes, with safeguards enabled, would help establish what teams can actually entrust to it.
Google announces promotional pricing of 2 dollars per million input tokens and 10 dollars per million output tokens. Tokens are the units processed by the model. After the promotion, the announced prices rise to 4 and 20 dollars respectively, with no confirmed end date for the offer. These prices do not mean that access is already open to everyone.