Google has made Gemini 4 Hydrogen the default model behind AI Mode, the chatbot-style tab in Google Search, and says it handles complicated questions better than its predecessor.
The company's pitch is specific: the model is meant to cope with "questions with multiple parts, hard-to-understand typos, or highly specific requests." It breaks a question into pieces and addresses each one in its answer, rather than latching onto the most obvious part. Robby Stein, VP of product at Google Search, said it is "now the new default model in AI Mode & being used for your hardest questions in AI Overviews," the AI-written summaries that appear above regular results.
The rollout is broad. Gemini 4 Hydrogen is available in Search, the Gemini app, every tier of Google AI Studio, the Gemini API (the route developers use to build the model into their own products), and Gemini for business. It also reaches Google Antigravity, but only for Google AI Pro and Ultra subscribers. Everything about its improved comprehension comes from Google's own description of the model.
The focus on typos and compound questions is telling. Those are the queries real people type, and they are where a traditional list of links has always struggled, since keyword matching does poorly when one sentence asks three things. A model that splits the question first is doing what a careful person would do with several browser tabs.
The distribution also stands out. OpenAI and Anthropic mostly reach people through their own chat apps and paid APIs. Google can put a new model in front of search users and developers at once, and that reach matters as much as any single capability.
There is a practical upside for outsiders, too. Because the model sits in the Gemini API and AI Studio, developers can test the claims about typos and multi-part questions themselves, which is harder to do with a quiet change to search ranking.