Google Search can now answer complex queries more accurately
What's the story
Google has announced a major update to its Search engine, powered by the new Gemini 4 Hydrogen model. The update is aimed at improving the search engine's ability to understand complex queries, including those with multiple parts or specific requests. The company describes this as its "biggest model improvement to Search yet."
Model details
The new model is designed for fast execution
The new Gemini 4 Hydrogen model is designed for fast execution across a variety of tasks.
It completes tasks 2.5 times faster than its predecessor, Gemini 3.8 Flash, while still being smart enough to outperform Gemini 4 Argon on the Artificial Analysis Intelligence Index at its default high thinking level.
This makes it an ideal choice for users who don't always need the most powerful model but prefer a faster and reasonably intelligent one.
Improved features
How the new model improves Google Search
The Gemini 4 Hydrogen model brings several improvements to Google Search, including a better understanding of user intent and more helpful links.
It can also provide visual responses and improved answers in different languages supported by Google's AI Mode.
The model is particularly good at handling nuanced questions, multiple instructions, and even confusing typos, making it easier for users to get the information they need without having to rephrase their queries repeatedly.
Technique enhancement
It also enhances Google's search fan-out technique
The Gemini 4 Hydrogen model also improves Google's search fan-out technique, which breaks down a single query into multiple sub-queries.
This helps users find relevant websites they might have otherwise missed.
The integration of this model into AI Mode will also improve the latter's visual capabilities, providing more images, structured information, comparison tables, and custom tools in response to user queries.
Model performance
The new Gemini model is also coming to AI Studio
Google claims that the Gemini 4 Hydrogen model can complete tasks faster than its predecessor, while still being strong on intelligence benchmarks.
The model offers four thinking levels: low, medium, high (default), and an extra-high tier ("xhigh").
It is also coming to all tiers of Google's AI Studio, the Gemini API, and the Gemini App for paid AI Pro and Ultra subscribers through Google's Antigravity development platform.