Google rolls out Gemini 4 Hydrogen, its new default AI Mode model for complex questions

Google ships Gemini 4 Hydrogen

Google has begun rolling out Gemini 4 Hydrogen, a new version of its Gemini model that the company says handles complicated questions better than its predecessors. According to The Verge, Google highlights the model's ability to deal with questions that have several parts, contain confusing typos, or make highly specific requests. Instead of treating such a query as a single block of text, the model breaks it apart and addresses each component in its answer.

The model is available across a broad set of Google products as of launch day, The Verge reports: Search, the Gemini app, every tier of Google AI Studio, the Gemini API, Gemini for business, and Google Antigravity for subscribers on Google AI Pro and Ultra plans. Robby Stein, VP of product at Google Search, said the model is now the default in AI Mode and is being used for the hardest questions that appear in AI Overviews.

Richer, more visual answers

Gemini 4 Hydrogen also changes what the answers look like. According to The Verge, the model runs simultaneous queries across the relevant subtopics of a question, which Google says produces a wider range of links in each response. Answers also carry more visual elements, including images, comparison tables, graphs, and maps.

One demonstration Stein shared, as described by The Verge, had the model generate an interactive animated driving simulation to explain how phones work out which highway lane a car occupies when there is no cell service. Google also claims improved language understanding across every language that AI Mode supports.

New reasoning dial and performance claims

Users get additional control over how much thinking the model does before answering. The Verge lists four levels: low, medium, high, and a newly added xhigh option.

On performance, Google claims the new model makes significant gains in speed and intelligence compared with Gemini Flash 3.8, and that it holds its own against, and in some cases beats, rival models in its class, naming Claude Sonnet 5.5 and GPT-6.1 Terra. These comparisons come from Google itself; The Verge's report does not include independent benchmark results, so the competitive claims remain unverified by third parties.

Why it matters

The most consequential detail is placement. By making Gemini 4 Hydrogen the default model in AI Mode and routing difficult AI Overviews queries to it, Google is putting its strongest question-handling model in front of its largest audience, Search users, without anyone opting in.

The focus on multi-part questions also targets one of the most common complaints about AI assistants: answers that quietly drop half of a compound request. If the decomposition behavior works as described, it changes how people can realistically use Search, allowing longer, messier and more specific prompts instead of a chain of separate lookups.

The launch also keeps pressure on Anthropic and OpenAI in the middle tier of the model market, and the xhigh thinking option extends the industry-wide move toward user-adjustable reasoning depth. As with any launch-day announcement, the speed and intelligence claims deserve skepticism until independent evaluations appear, but the distribution alone makes this a significant release to watch.