Google's Gemini 4 Hydrogen upgrades Search's answers

Google is rolling out its new Gemini 4 Hydrogen model across Search, which the company calls its “biggest model improvement to Search yet.” Unlike Google’s frontier models, Gemini 4 Carbon and Gemini 4 Argon, which are tuned for deep reasoning on hard problems, Hydrogen is built for fast execution across agentic workflows, where an AI completes several steps on its own without additional instructions — on the theory that the most powerful model isn’t always the right one for the job. Google says Hydrogen completes tasks 2.5 times faster than its previous fast model, Gemini 3.8 Flash, while still beating the more powerful Argon on a wide-ranging intelligence benchmark — the Artificial Analysis Intelligence Index — when running at its default high-thinking setting.

In Search, Hydrogen is meant to better parse multi-part, highly specific or typo-ridden questions, letting people ask in their natural train of thought rather than simplifying for the engine. Its improved reasoning also strengthens Search’s “fan-out” technique, which splits a single query into multiple subqueries so users surface related sites they might otherwise miss, and brings more images, comparison tables and other visual responses into AI Mode. Google says Hydrogen can also work independently on everyday coding tasks — writing code, running tests, verifying changes — or in combination with Carbon and Argon in multi-model workflows, and that it offers four thinking levels, including a new “xhigh” tier, all aimed at finishing tasks faster using fewer output tokens. On multi-step work such as deep research and codebase navigation, the company says Hydrogen ranks fifth on Agent Arena’s benchmarks.

Beyond Search, Gemini 4 Hydrogen is rolling out to every tier of Google’s AI Studio, the Gemini API and the Gemini App, as well as to paying AI Pro and Ultra subscribers on Google’s Antigravity development platform; the Search update itself is reaching all users globally.

Why it matters

By optimizing a model for speed and agentic tasks rather than raw reasoning power, Google is betting that most everyday queries — in Search and beyond — are better served by a faster, cheaper model than by routing everything through its most expensive frontier systems.