Google announces Gemini 4 Carbon, releasing its new frontier AI model to cyber defenders first

Google on Wednesday 28 October announced Gemini 4 Carbon, which it called its “next frontier model” and its strongest so far for cybersecurity and long, multi-step software engineering work. The model is not yet generally available.

Koray Kavukcuoglu, senior vice president at Google DeepMind and Google’s chief AI architect, wrote that Google is taking “a deliberate approach to rollout”. Starting this week, Carbon is being made available to a subset of cyber defenders, with access to be extended gradually to other trusted partners through Google’s Fairwind Program. Google DeepMind says the programme gives priority to governments, critical-infrastructure operators and core technology platforms, and limits access to their internal security, incident-response and penetration-testing teams. Google said it would release the model to developers, businesses and consumers “as soon as possible” after gathering feedback and refining its guardrails, without giving a date. It also said it is taking part in the US government’s voluntary process for testing models before release.

Google’s figures. The benchmark results below are Google’s own, with the cybersecurity tests run without guardrails. Google said Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59, and scored 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploiting them) and 71.0% on CWE-bench (patching them). It said Carbon has run for days on internal engineering tasks and that, in internal testing, it made about half as many unsupported claims as Gemini 4 Argon, released two weeks earlier.

Google said the model will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 when the introductory period ends.

See also: Google DeepMind releases Genie 4 world model with real-time audio and mid-simulation prompts.