Source: www.theverge.com/ai-artificial-intelligence/1032505/google-search-ai-mode-gemini-4-hydrogen
Google says the model can split a question into parts and address each one in its answer. It runs simultaneous queries across the relevant subtopics, which it says widens the range of links in responses. Responses also include more visuals, such as images, comparison tables, graphs and maps.
Robby Stein, VP of product at Google Search, said the model is now the default in AI Mode and is used for harder questions in AI Overviews. Google also says language understanding improved across all supported AI Mode languages. Users get more thinking levels, including low, medium, high and a new xhigh setting.
Google says the model brings significant gains in speed and intelligence over Gemini Flash 3.8. It also says Gemini 4 Hydrogen matches or outperforms some competing models in its class, including Claude Sonnet 5.5 and GPT-6.1 Terra. These comparisons are Google's own claims.
Why it matters if you build agents
Developers using the Gemini API or Google AI Studio can now test a model with a new xhigh thinking level alongside the existing settings. Agents that field multi-part or messy user queries may need fewer prompt-splitting workarounds, though the announcement gives no pricing or latency figures. Teams should check the performance claims against their own evaluations before changing defaults.
Get stories like this every weekday
One short email with the day's agent tooling news.