On 28 October 2026 Google DeepMind announced Gemini 4 Carbon, which Google calls its "strongest model with state-of-the-art cybersecurity and long horizon capabilities". It is built on Gemini 4 Argon and arrived about two weeks after Argon. Google says its own teams have watched Carbon work for days on one task, getting past dead ends to find and fix the root causes of subtle bugs in large codebases and low-level systems.
Security results, all reported by Google: 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (proving exploitability) and 71.0% on CWE-bench (patching). Google says these security tests ran without guardrails. In internal tests Carbon generated secure code successfully more than twice as often as Argon. Google says it ranks first on the Artificial Analysis Intelligence Index v5.0, with a score of 59, and fourth on WebDev Arena. It also reports a 0.0% attack success rate on Gray Swan's indirect prompt injection benchmark, about half as many unsupported claims as Argon, and almost 80% fewer high-severity misalignment alerts in internal coding environments. Apart from the Artificial Analysis ranking, these figures have not been independently checked.
Who can use it: from launch week, only a subset of cyber defenders can use Carbon. Access will widen gradually to other trusted partners through Google DeepMind's Fairwind Program, which ETIH says prioritises governments, healthcare providers, telecoms and others defending critical systems. Google is also taking part in the US government's voluntary pre-release model testing. It says developers, businesses and consumers will get Carbon "as soon as possible" but has given no date. Commentators link the gated launch to dual-use risk: the skills that help defenders could also help attackers.
Pricing: during an introductory period, Carbon costs the same as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After that, Google says the price rises to $4 and $20.
