By Hamza Faisal

Google DeepMind has unveiled Gemini 4 Carbon, a new frontier AI model it calls its strongest yet, built around cybersecurity and long-running engineering work. Announced on October 28, it arrives just two weeks after its predecessor, Gemini 4 Argon — a pace that says as much about the frontier race as the model itself.

Carbon ranks first on the newly updated v5.0 of the Artificial Analysis Intelligence Index with a score of 59, according to Google, ahead of Anthropic's Claude Opus 5.5 at 58, OpenAI's GPT-6 Astra at 57 and Argon itself at 50. Google DeepMind senior vice president Koray Kavukcuoglu described it as the company's "strongest model with state-of-the-art cybersecurity and long horizon capabilities."

Built for the full defense cycle

Cybersecurity is the headline. Google says Carbon leads across the whole defense cycle: finding vulnerabilities, testing whether they can be exploited, and writing the fix. It scored 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation testing and 71.0% on CWE-bench for patching — all, Google says, state-of-the-art results.

The internal claims go further. Carbon generates secure code at more than twice the success rate of Argon, Google said, and its Chrome team reported that it performed more than 10 times better than another lab's large frontier model when generating high-quality patches — though Google did not name the rival. The model also recorded a 0.0% attack-success rate on Gray Swan's indirect prompt injection benchmark, and internal teams using it in live coding environments reported a nearly 80% drop in misalignment alerts. On the web development side, it gained 89 Elo points on WebDev Arena to reach 1769, good for fourth place.

A deliberately slow rollout

For all the benchmark fanfare, almost nobody can use Carbon yet. Google is taking what it calls a deliberate approach: starting this week, access goes to a subset of cyber defenders, then gradually to additional trusted partners through its Fairwind Program, which spans more than 650 partners including governments, national cyber authorities and critical infrastructure operators in healthcare, energy, telecom and finance. The company says it is also participating in the US government's voluntary process for pre-release model testing.

Developers, businesses and consumers will get access only after Google gathers feedback and iterates on guardrails — "as soon as possible," in its words, with no date given. The caution reflects the dual-use tension at the heart of the model: a system that can find and fix flaws can also, in the wrong hands, help exploit them. Google says it is testing that tradeoff before a wider release.

The frontier race, two weeks at a time

Carbon launches at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input priced 95% off — rates that step up to $4 and $20 once the introductory window ends. The launch landed the same day Anthropic's Frontier Red Team said an unreleased Claude model had chained flaws into working browser and operating-system exploits, a reminder that cyber-defense capability is where the labs are now fighting out the frontier.

Some caution is warranted: the headline numbers are Google's own and have not been independently reproduced, and independent testing found Carbon trailing rivals on some professional tasks while recording a higher hallucination rate than Argon on one factual-knowledge test. Still, two flagship launches in fourteen days show how fast the frontier is moving — and the open question is whether defenders get tools like Carbon before attackers match them.