Google today made Gemini 4 Hydrogen, the third model in its Gemini 4 line, available in Google Search for all users globally, thirty days after it put the first one, Gemini 4 Argon, in Search only for paying subscribers. Robby Stein, Vice President of Product for Google Search, called it "the biggest model upgrade to Search yet" in a short blog post that lists four improvements - intent, links, visual answers and language - and contains no figures, no list of the Search surfaces involved and no mention of advertising.

In Short

Google put a new AI model, Gemini 4 Hydrogen, into Google Search today, and unlike the two Gemini 4 models before it, this one is open to everyone rather than only to people paying for Google's AI plans. Google says it handles complicated questions better, runs more searches across subtopics to surface a wider range of websites, and answers with more images, comparison tables and custom tools, all of which bear on which sites get cited and how much room is left for links and ads. The post gives no numbers for any of it and says nothing about ads, clicks to websites or what happens to the paid Gemini 4 models already in Search.

A two-minute post with no numbers

The announcement is brief. Titled "Google Search gets a major upgrade with Gemini 4 Hydrogen," dated today and labelled a two-minute read, it sits in the Search section of Google's blog under Stein's byline. It opens: "Today, Google Search is bringing cutting-edge intelligence to everyone with Gemini 4 Hydrogen. This is the biggest model upgrade to Search yet."

The second paragraph carries the substance. "Hydrogen is now available globally in Google Search for all users, delivering more helpful, visual, and intuitive responses with improved model capabilities," according to the post. Four bullet points follow, on intent, on query fan-out and linking, on visual responses and on language. The post closes with a single line: "Try out Gemini 4 Hydrogen in Search today."

What the post leaves unsaid is as important as what it contains. It refers throughout to "Search" in general. AI Mode appears once, in a sentence about languages, and AI Overviews, the AI summaries that sit at the top of ordinary results pages, are not mentioned at all. The post does not say which model Hydrogen replaces, whether it now writes every AI response in Search or only some, whether the rollout is complete in every country, or what changes for subscribers who already choose between models in AI Mode. There is no benchmark, no latency figure, no count of links and no traffic data.

Superlatives are not new from Google Search either. At I/O on May 19, 2026, Google described its redesigned query box as the biggest upgrade to its Search box in over 25 years. Today's claim is narrower - the biggest model upgrade - and the post offers nothing against which to measure it.

Three Gemini 4 models, three audiences

The distribution is the clearest change. When Google released Argon on October 13, consumers could reach it only on the paid AI Pro and Ultra plans, in the Gemini app and in AI Mode, and developers paid an introductory $2 per million input tokens and $10 per million output tokens.

Carbon, the second Gemini 4 model, narrowed the audience further. Sundar Pichai introduced Carbon on October 28 as Google's strongest frontier model, saying it ranked first on the Artificial Analysis Intelligence Index with a score of 59. On November 10, Google's Rajan Patel said Carbon was rolling out in AI Mode to Google AI Ultra subscribers, with a model menu in the search box offering Auto, 3.8 Flash and 4 Carbon. Ultra is a plan PPC Land has reported starting at $99.99 a month.

Two days later, the third model goes to everyone. In thirty days, Google has moved from a Gemini 4 model in Search for paying subscribers only to one available, in its words, to "all users."

The everyday model in Search has changed several times this year. At I/O in May, Google replaced the previous default in AI Mode with Gemini 3.5 Flash. AI Mode then moved to Gemini 3.8 Flash on September 2. The split between a fast model for the mass audience and a stronger one for harder questions goes back further: by late December 2025, Gemini 3 Flash powered the Search integration for most users, while Google routed challenging questions in AI Mode and AI Overviews to Gemini 3 Pro.

Today's Search post does not name the model Hydrogen takes over from. Google's separate model announcement, published the same day, positions Hydrogen as the upgrade for teams moving off Gemini 3.8 Flash, the model that AI Mode has used since September.

What Hydrogen is, according to Google

That companion post, written by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, carries the technical detail the Search post omits. Google describes Hydrogen as a fast, cost-efficient model built for agentic workflows and everyday tasks. Carbon and Argon, in Google's framing, handle deep, long-horizon reasoning on the hardest problems, while Hydrogen is built for rapid execution and can run on its own or alongside the two larger models.

The developer price places it in the Flash tier. According to Google, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, flat across its 1 million-token context window. Those are the rates Google set in July, when it priced Gemini 3.6 Flash at $1.50 input and $7.50 output. Against Argon's introductory $2 and $10, both of Hydrogen's rates are 25% lower, by PPC Land's calculation.

Google also claims Hydrogen performs close to Argon. According to the model post, Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting, against 50 for Argon, and 54 at its highest setting; Carbon's score, as Pichai gave it in October, was 59. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at the default setting and cuts fabricated claims, or hallucinations, by more than five times compared with that model. The model offers four thinking levels - low, medium, high and a new extra-high setting - that trade quality against speed and cost.

All of these figures come from Google or from evaluations it selected. None has been reproduced independently, and the Search post does not say which thinking level Search uses, or whether it varies by query.

Pichai hinted at a model of this kind in October. In an interview published on October 21, he said Google remained capacity constrained after Argon's launch and that more efficient versions, nearly as capable, would follow, without giving a date. Google's posts do not describe how Hydrogen was built or whether it is derived from Argon. A model priced at Flash rates, scored at Argon's level and released to the whole Search audience three weeks later matches the profile Pichai sketched, though Google has not drawn that connection.

The Search rollout is also broader than the Gemini app rollout. In the app, according to a footnote in the model post, free users reach Hydrogen only through automatic model selection, when the system judges it appropriate for a prompt. Free Gemini accounts have been limited to Flash-Lite since October 9. Search, the product that carries Google's advertising, gets the wider release.

Query fan-out, upgraded again

For publishers, the second bullet point matters most. "Hydrogen's advanced reasoning supercharges our query fan-out technique in Search, connecting you to a broader range of helpful links than before," the post states. It continues: "By more thoroughly understanding your question and conducting simultaneous queries across all the relevant subtopics, Search helps you find sites you previously might not have discovered."

Query fan-out is the mechanism behind AI Mode's answers. Google has described it as a way to divide a user query into multiple subtopics and issue searches for each simultaneously, and Nick Fox spoke in May of potentially hundreds or thousands of sub-queries issued at once. The results are synthesised into one response with links.

This is not the first model upgrade sold on fan-out. In April 2025, with Stein already VP of Product for Search, Google applied the technique to image searches, issuing multiple queries about a picture and the objects in it. In September 2025, PPC Land reported AI Mode breaking questions into subtopics while processing hundreds of related searches. When Gemini 3 arrived on November 18, 2025, Google said its reasoning let query fan-out perform more searches uncovering relevant web content. Each model generation has come with a similar promise of wider coverage.

What has rarely come with it is measurement. Today's post gives no count of sub-queries, no average number of links per answer and no indication of which kinds of sites benefit from the "broader range." It also does not say where the extra links appear. In May, Google added five outbound link features to AI Mode and AI Overviews, among them an "explore new angles" section, labels for news subscriptions, a first-hand perspectives section, inline links and desktop hover previews. Whether Hydrogen changes how those placements are filled is not addressed.

A longer list of sources is also not the same as more visits. On October 29, Reddit chief executive Steve Huffman told analysts that AI Mode does not drive traffic, to Reddit or, as far as he knew, to anyone else, as Reddit warned that its 130 million daily active uniques could fall in the fourth quarter. Google's post offers no click or referral data that would test either claim.

The third bullet point describes "highly adaptive, visual, and easy-to-scan responses that are uniquely relevant to your needs," delivered "at the speed of Search." According to the post, "You'll see more images, structured responses, comparison tables, and custom tools built for you."

Custom tools have a history in Search. With Gemini 3, Google introduced generative interfaces in AI Mode, among them a loan calculator built inside the answer for someone researching mortgages, initially for US subscribers and with generation times that could exceed a minute. By December 2025, the model was writing code for simulations, visualisations and widgets that run directly inside search results. At I/O in May, Google said these custom interfaces would reach everyone in Search over the summer, free of charge.

The phrase "at the speed of Search" is the notable addition. Earlier generative layouts could take more than a minute; today's post claims visual answers without a wait, though it gives no timing.

Comparison tables and custom tools are formats that can complete a task inside the answer - weighing products, running a calculation, laying out options side by side. Those are also tasks that review sites, comparison publishers and retailers have built businesses around. The post does not say how visual elements sit alongside links, or alongside ads.

Intent, typos and tool calling

The first bullet point is about understanding. According to Google, "Search can now handle even more nuanced and complicated questions, quickly understanding what you need," including "questions with multiple parts, hard-to-understand typos, or highly specific requests." The post adds: "With deeper instruction following and tool calling, Search more precisely follows what you want and gives you relevant responses that address each part of your question."

It does not say which tools Search calls. Two days earlier, Patel had described Carbon in AI Mode in similar terms, saying it understood intent better and allowed deeper back-and-forth conversations. Longer, multi-part queries are where AI Mode already lives: Google has said the average AI Mode query is three times the length of a traditional search query, and that AI Mode passed one billion monthly users in May.

Those longer queries are also where advertisers have the least control. When Google extended ads in AI Mode to ten more countries on October 28, its documentation said exact and phrase match keywords are limited to direct-intent queries there. The post does not say whether a model that reads multi-part questions more closely changes which ads qualify for them.

Languages without a list

The fourth bullet point reads: "No matter what language you ask in, you'll now get more highly nuanced and helpful responses across all supported AI Mode languages." The post does not list those languages or give their number.

AI Mode has added languages in batches. In September 2025, Google added Hindi, Indonesian, Japanese, Korean and Portuguese. The advertising side has not kept pace: the Google help page covering ads in AI Overviews and AI Mode lists English as the only supported language, according to PPC Land's coverage of the October 28 expansion. Answers improve in every AI Mode language from today, on Google's account; ads in those answers remain an English-language product.

The missing subject: money

Advertising appears nowhere in the post. That absence stands out against the past three weeks.

On October 28, Google scaled ads in AI Mode to Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines and Singapore, with Brazil, Mexico and the United Kingdom planned for the fourth quarter. Existing Search, Shopping and Performance Max campaigns are eligible automatically, advertisers cannot opt out of AI Mode separately from Search, and Google can wrap ads in generated "AI explainers" that advertisers cannot write, edit, approve or switch off. On November 4, Search Engine Roundtable documented up to four ads beneath a single AI Mode answer, where one to three had been usual.

The same day ads expanded, Alphabet told investors that strong growth in AI Mode queries had weighed on results because of lower monetisation. Class A shares fell 12.6% on October 29, their worst day since the company listed in 2004.

Hydrogen puts the economics in sharper relief. A more capable free model, with richer visual answers, is designed to make AI responses more useful. If that draws more queries into AI Mode, it extends the part of Search that Alphabet itself said earns less per query. The post does not address that trade-off, and nor does it say what Hydrogen costs Google to serve at Search scale; the developer price is a list price, not an internal cost.

Timing adds a complication for anyone tracking search visibility. Google's November 2026 core update began on November 10 and may take up to two weeks to complete, which puts its end close to the Black Friday trading period. Hydrogen's arrival two days into that rollout means two changes - one to rankings, one to the model that selects and summarises sources - are now running at the same time. Separating their effects on traffic will be difficult.

Timeline

Summary

Who: Google, in a post by Robby Stein, Vice President of Product for Google Search, with technical detail in a companion post by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google. The change affects every Google Search user, and with them the publishers whose pages are cited in AI answers and the advertisers whose ads appear alongside them.

What: Gemini 4 Hydrogen, the third Gemini 4 model, became available in Google Search for all users. Google says it improves understanding of complex questions, widens query fan-out to link a broader range of sites, adds more images, comparison tables and custom tools to answers, and improves responses in all AI Mode languages. The post gives no figures and does not mention ads. For developers, Google prices Hydrogen at $1.50 per million input tokens and $7.50 per million output tokens, the same as Gemini 3.6 Flash.

When: Today, thirty days after Argon reached AI Mode for paying subscribers on October 13 and two days after Carbon began rolling out to Ultra subscribers on November 10.

Where: Google Search globally. The post names AI Mode only in connection with languages, does not mention AI Overviews and lists no countries.

Why: Google is moving a Gemini 4 model from its paid tiers to the full Search audience. For marketers, that changes the model choosing and summarising sources for every Search user, during a core update and weeks after Alphabet said AI Mode growth had weighed on search monetisation, while Google has published nothing yet on how the change affects links, clicks or ads.