If you searched for anything on Google this weekend, a new model probably answered you. Google has put Gemini 4 Hydrogen behind its search engine, calling it the biggest model upgrade to Search yet. The change went live Thursday, and by Saturday the rollout had become one of the most searched tech stories in the United States.
Robby Stein, Google's vice president of product for Search, announced the switch on the company's Keyword blog on November 12. Hydrogen is now the default model in AI Mode, Google's chat-style search experience, and it handles the hardest questions behind AI Overviews, the AI summaries at the top of a results page. Google says it is available globally in Search for all users, with no waitlist.
The pitch is aimed at the questions that used to trip search up. According to Google, Hydrogen can take a question with several parts, a typo that scrambles the meaning, or a request so specific it barely looks like a search query, and still work out what the person actually wants. It breaks the question into its components and answers each one, rather than guessing at the gist. Stein's example was an interactive driving simulation built into an answer — one of the more visual responses Google says are coming.
Under the hood, the change leans on a technique Google calls query fan-out. One question gets split into several smaller searches run in parallel, and the results get stitched back together. Fan-out is not new. What changed is the model running it. Google says Hydrogen's reasoning sharpens the technique, so AI Mode should link to a broader range of websites than before. The company published no figures for either claim.
Hydrogen is the fast, cheap member of Google's new Gemini 4 family, alongside Carbon, the paid model that arrived November 10, and Argon at the top of the range. It replaces Gemini 3.8 Flash, the model that previously powered the quick answers in Search. Google claims it completes tasks two and a half times as fast as 3.8 Flash in the company's own testing. Those are Google's numbers. No independent benchmark results were included in the announcement, and no outside lab has published its own tests yet.
It has four thinking levels: low, medium, high (the default), and a new xhigh for the hardest problems. For developers using it through the Gemini API, Google set the price at $1.50 per million input tokens and $7.50 per million output tokens, flat across a one-million-token context window. In the Gemini app, Hydrogen is the model free users get through automatic selection.
The rollout goes well beyond Search — Hydrogen is now in the Gemini app, every tier of Google AI Studio, the Gemini API, and Gemini for business. Google is not launching a standalone product. It is swapping the engine inside services billions of people already use.
What it means depends on what you ask. A simple question — a score, a time, a definition — will look the same as last week. The difference should show up in the messy ones: a trip with three conflicting constraints, a comparison between products with different specs, a question typed in a hurry with half the words misspelled.
There is a catch. A stronger model does not mean AI Overviews stop making mistakes. The promise of more links is Google's claim, not a measured result, and the company has not said when every account will see the change. For now, the upgrade is live, the searches are surging, and the test is happening in public, one query at a time.



No comments:
Post a Comment