Gemini 4 Carbon: What Google's New Frontier Model Changes

Two weeks after Gemini 4 Argon, Google is already moving the frontier again. On October 28, 2026, Google DeepMind announced Gemini 4 Carbon, which the company calls its strongest model yet, with standout results in cybersecurity and in long-horizon agentic work. Rather than a lighter, cheaper sibling, Gemini 4 Carbon is a step up from Argon at the same introductory price, and Google is releasing it in deliberate stages, starting with trusted cyber defenders.

• Announced: October 28, 2026, by Koray Kavukcuoglu, SVP at Google DeepMind and Google's Chief AI Architect
• Benchmarks (Google's figures): 59 on the Artificial Analysis Intelligence Index v5.0 (ranked first), 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench
• Introductory pricing: $2 per million input tokens and $10 per million output tokens, with cached input 95% off; The Register reports it rises to $4 and $20 after the introductory period
• Access: a subset of cyber defenders first, then trusted partners through Google's Fairwind Program, with broader developer and consumer availability to follow

What Is Gemini 4 Carbon Built to Do?

Google frames Carbon around two strengths. The first is sustained, long-horizon engineering. According to the announcement, internal teams have watched Carbon work for days on a single complex task, such as root-causing and patching subtle bugs across large codebases and low-level systems, persisting through dead ends rather than giving up. Google also reports gains in full-stack web development, scientific tasks and multimodal understanding, with Carbon climbing 89 Elo points to fourth place on WebDev Arena.

The second strength is defensive security, and this is where the frontier model stands apart. Google describes state-of-the-art results across the full defense lifecycle: finding vulnerabilities, validating which ones are actually exploitable, and writing root-cause fixes. On internal tests, Google says Carbon more than doubles Argon's success rate at generating secure code, and the Chrome team reported Carbon producing high-quality patches more than 10 times as effectively as another lab's large frontier model. Google notes the external benchmark scores were evaluated without guardrails.

How Does Carbon Compare With Argon and Rival Models?

Independent coverage puts these claims in useful perspective. Ars Technica observed that Carbon edges Argon in every test Google published and narrowly leads rival frontier models on most of them, while noting that every frontier system now scores within a few points of the others. The Register reports that Artificial Analysis places Carbon first on its new Intelligence Index v5 and second on its Coding Agent Index, and that Carbon ranks second on Arena.ai's Best Overall Agent leaderboard. In other words, the leading models are tightly bunched, and this release gives Google a strong claim to the top spot in general reasoning and security tasks.

Safety work features prominently too. Google says Carbon cut successful cheating on Terminal-Bench 4.0 to 0.0%, reduced high-severity misalignment alerts in live internal coding by almost 80%, and roughly halved unsupported claims compared with Argon. Ars Technica highlighted that Carbon resisted every attempted prompt-injection attack in the Grey Swan evaluation.

Why a Staged Rollout Makes Sense for a Cyber-Capable Model

A model that is very good at finding and exploiting vulnerabilities is enormously valuable to defenders, and Google is choosing to put it in their hands first. Carbon is going to a subset of cyber defenders this week, then to more trusted partners through the Fairwind Program, which SecurityWeek reports includes more than 650 organizations. Google says it is also participating in the US government's voluntary pre-release model testing process. That mirrors the path Argon followed, which we covered when Gemini 4 Argon reached cyber defenders first before Argon's general availability and pricing.

For developers, the practical takeaway is pricing parity: when Carbon reaches the Gemini API, it starts at the same introductory rates as Argon, so teams can upgrade without rebudgeting. Google has not yet given a date for wider availability, only that it will come as soon as possible.

The bigger picture is the pace. Alphabet CEO Sundar Pichai said in the company's earnings remarks that its first-party model APIs now process nearly 32 billion tokens per minute, more than four times the volume of a year ago. With Argon and Carbon landing a fortnight apart, Google DeepMind is shipping frontier upgrades at a remarkable clip. Follow every release in our AI coverage.

Sources: Google Blog — October 28, 2026; Ars Technica — October 28, 2026; The Register — October 29, 2026; SecurityWeek — October 29, 2026.