Google rolls out Gemini 4 Hydrogen to Search for all users globally

Google Search is now powered by Gemini 4 Hydrogen for all users worldwide, Google announced on November 12, 2026. In a post by Robby Stein, VP of Product for Google Search, the company calls it "the biggest model upgrade to Search yet." That characterization is Google's own.

What Google says changed

Google lists four areas of improvement.

Intent understanding. Google says Search can now handle more nuanced and complicated questions, including multi-part questions, hard-to-understand typos and highly specific requests. It attributes this to "deeper instruction following and tool calling," which it says lets Search address each part of a query more precisely.

Query fan-out and linking. Google says Hydrogen's reasoning improves its query fan-out technique, in which Search runs simultaneous queries across the relevant subtopics of a question. According to Google, this connects users to a broader range of links than before, including sites they might not previously have found.

Visual responses. Google says answers are now more adaptive and easier to scan, with more images, structured responses, comparison tables and custom tools built for the query. The company says these arrive "at the speed of Search."

Language understanding. Google says responses are more nuanced and helpful across all supported AI Mode languages.

Availability

Hydrogen is available now in Google Search globally, for all users. Google's examples of use cases include travel, shopping and hobby queries with specific requirements phrased in natural language.

Technical details

Google's announcement does not give a parameter count, context window, benchmark scores or API pricing for Gemini 4 Hydrogen. Those figures are not yet disclosed. All performance statements in the post are Google's claims, and no independent evaluation accompanies them.

What this means

The announcement is a deployment of a new Gemini model into Google's main consumer search product, with availability for all users rather than a limited preview. The emphasis on instruction following and tool calling indicates that Google is tying Search quality to the model's ability to break a question into parts and act on each one. The query fan-out description points the same way: more simultaneous sub-queries, informed by the model's reasoning, to surface a wider set of links.

The move toward tables, images and custom tools also changes what a Search result looks like for complex queries. Publishers and developers will want to see how the broader link range Google describes plays out in practice. Measured results on accuracy, latency and link diversity are not yet available, so Google's claims remain unverified.