Google has announced Gemini4Hydrogen and is bringing it to Google Search worldwide, with a focus on improving the search engine's ability to handle multi-part queries, complex details, and misspellings. Google says this is the largest model upgrade to Search to date.
Unlike Gemini4Carbon and Gemini4Argon, which focus on complex reasoning, Hydrogen is mainly designed for high-speed agentic workflows and can autonomously complete multi-step tasks. Google says it executes tasks 2.5 times faster than the previous-generation fast model, Gemini3.8Flash. At the default high thinking level, Hydrogen even outperforms the more capable Gemini4Argon on the Artificial Analysis Intelligence Index benchmark.
In search scenarios, Hydrogen can more accurately understand user intent, follow instructions, and invoke tools. It also improves Search's fan-out technique by splitting complex prompts into multiple subqueries to discover more relevant websites. Google AI Mode will also provide more images, structured answers, comparison tables, and customized tools, further improving the multilingual search experience.
Hydrogen supports four thinking levels: low, medium, high, and a newly added extra-high (xhigh) level, aimed at reducing output token consumption and improving task execution efficiency. On the Agent Arena benchmark, it ranks fifth in multi-step tasks such as deep research, data analysis, and large codebase navigation. In addition, the model can autonomously write code, run tests, and verify code changes, and can also collaborate with Carbon and Argon to complete workflows.
In addition to Google Search, Gemini4Hydrogen will also be available on Google AI Studio, the Gemini API, the Gemini App, and the Google Antigravity development platform for AI Pro and Ultra paid subscribers. This upgrade shows that Google is further advancing search from traditional information retrieval toward AI agents capable of executing complex tasks.
