Google just rolled out Gemini 4 Hydrogen, a new model built for complicated questions. According to The Verge AI, it’s designed for “questions with multiple parts, hard-to-understand typos, or highly specific requests.” It splits your question into parts and answers each one.
The rollout is wide from day one. Robby Stein, VP of product at Google Search, says the model is “now the new default model in AI Mode & being used for your hardest questions in AI Overviews.” So millions of people will use it without changing a single setting.
🚀 What Gemini 4 Hydrogen Brings
- It answers every part of a multi-part question. Ask something with three moving parts and older models often handle one or two and skip the rest. Hydrogen is built to split your question into pieces and work through each one. It’s also better at figuring out what you meant when your query is full of typos.
- You get a wider range of sources. Google says the model runs “simultaneous queries across all the relevant subtopics.” Instead of one search, it fans out across every angle of your question, so you see more links. For publishers, that could spread traffic across more sites, although Google hasn’t said how this affects click-through.
- Answers include more visuals. Responses now have more images, comparison tables, graphs and maps. Stein’s demo goes further. Someone asked how phones work out which highway lane a car is in when there’s no cell service, and the model built an interactive animated driving simulation to explain it. That’s a big step up from a block of text.
- Better language understanding. Google says language understanding improved “across all supported AI Mode languages.” That matters for the huge share of Search users who don’t search in English.
- A new “xhigh” thinking level. You can pick low, medium, high, and now “xhigh” thinking levels. Higher levels let the model reason longer before it answers, which usually means slower but more careful responses. Developers can trade speed for depth depending on the task.
📊 How It Compares
Google says Hydrogen “delivers significant leaps in both speed and intelligence” over Gemini Flash 3.8. It also says the model “matches or outperforms some of the competing models in its class,” and names Claude Sonnet 5.5 and GPT-6.1 Terra.
Read that line carefully. “Some of the competing models” is a hedge, and these are Google’s own claims. The article doesn’t include benchmark numbers or independent testing. Hydrogen is also positioned against mid-tier models, not the flagships from Anthropic and OpenAI. Expect third-party evaluations to show where it actually lands.
🔓 Who Can Use It
Gemini 4 Hydrogen is available starting today in:
- Google Search (AI Mode and AI Overviews)
- The Gemini app
- All tiers of Google AI Studio
- The Gemini API
- Gemini for business
- Google Antigravity, for Google AI Pro and Ultra subscribers only
The Antigravity restriction is the main catch for developers. The API and AI Studio are open across tiers, but the agentic coding environment sits behind a paid subscription. The article doesn’t mention API pricing.
🧭 Why This Matters
What stands out here is the distribution. Most model launches start in an API and slowly reach consumers. Google is putting Hydrogen into Search on day one, which gives it more real-world use on day one than most rivals get in months.
It also shows where AI search is heading. Answers that fan out across subtopics, pull in more sources and generate interactive visuals are moving Search away from ten blue links and toward a research assistant. The lane-detection simulation is a good example: Google is betting people want to be shown things, not just told.
The open questions are accuracy and trust. More visuals and more sources only help if they’re correct, and AI Overviews has had public stumbles before. Putting a new model on “your hardest questions” raises the stakes when it gets something wrong.
Expect OpenAI and Anthropic to answer quickly, since Google named their mid-tier models directly. Full details are available in The Verge AI’s original report.