Google has unveiled Gemini 4 Carbon, which it calls its strongest model yet, built for cyber defence and for coding tasks that run for days. The catch: almost nobody can use it. On 28 October 2026, Google said access begins with a small group of vetted cyber defenders, with no date yet for developers, businesses or consumers.
What Google announced
Koray Kavukcuoglu, who leads Google DeepMind, said Carbon is state of the art in cybersecurity and long-horizon work, meaning tasks that need sustained effort. Google says internal teams watched it keep going through dead ends for days while tracing and patching subtle bugs in large codebases. It arrives roughly two weeks after Gemini 4 Argon, which it succeeds.
Google frames Carbon around the full defence cycle: finding a flaw, validating that it can be exploited, and fixing it. According to Google, the Chrome team found Carbon more than ten times better than another lab's model at producing high-quality patches. Reported pricing is $2 per million input tokens and $10 per million output tokens, the same as Argon.
Who gets it first
Access runs through Google DeepMind's Fairwind Program, which gives high-priority defenders such as governments, healthcare providers and telecoms providers early access to advanced models. Google says it works with more than 650 partners. Participants must:
- pass background checks and use phishing-resistant multi-factor authentication;
- limit use to internal security, incident-response or penetration-testing teams;
- track who uses the model, and never share or sell access.
Google says feedback from this first group will help it refine safeguards before access widens "soon."
The benchmark claims
Coverage of Google's numbers reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, plus a first-place ranking on version 5 of the Artificial Analysis Intelligence Index with a score of 59. These figures come from Google's announcement. Because there is no public API, no downloadable model and no outside testing yet, nobody independent has verified them.
Why the staged release matters
A model that is good at finding and exploiting software flaws is useful to defenders and attackers alike. Google's own threat researchers have documented state-backed groups from Iran, North Korea, Russia and China experimenting with Gemini, mostly for research and coding help. Releasing to vetted defenders first is meant to give them a head start. It also reflects a growing pattern among AI labs: the most capable security models reach the public later, if at all.
The practical effect for most readers is that Carbon is a headline, not yet a tool. Security teams at governments, hospitals and telecoms operators can look at Fairwind's application channel. Everyone else will need to wait for Google to announce the next stage of access.
Sources: Google DeepMind announcement (28 October 2026); Unite.AI; MarkTechPost; Technology.org; Crypto Briefing; MindStudio.
