# Which Gemini model powers Google AI Mode in 2026?

> As of 17 November 2026, Google AI Mode uses Gemini 4 Hydrogen as its new default model.

Source: https://generativesearchnews.com/gemini-3-in-google-search · Publisher: Generative Search News · Published: 2026-11-17 · Updated: 2026-11-17

## Why it matters
- Google AI Mode’s default model is Gemini 4 Hydrogen as of 12 November 2026.[1]
- Google says Hydrogen handles the hardest AI Overview questions, but it has not published the complete model allocation for all Overviews.[1]
- Gemini 4 Argon and Gemini 4 Carbon provide separate subscriber-accessible model options in AI Mode.[5]
- Query fan-out means Google generates related searches across subtopics before assembling an answer.[10][15]
- Google’s Deep Search documentation still names Gemini 2.5 Pro and does not confirm a Hydrogen replacement.[1][9]

As of **17 November 2026**, Google AI Mode uses **Gemini 4 Hydrogen as its new default model**. Google also says Hydrogen handles users’ hardest questions in AI Overviews, while separate documentation identifies Gemini 2.5 Pro for Deep Search; Google has not published a complete model allocation for every AI Overview query.[1]

## What happened on 12 November 2026?

Google Search announced a major model upgrade centred on Gemini 4 Hydrogen. The model is available globally in Search for all users and is now the default model in AI Mode.[2][1]

Google’s wording for AI Overviews is narrower: Hydrogen is used for the hardest questions. That does not establish that every AI Overview runs on Hydrogen, nor does it disclose which model handles the remaining questions.[1]

The November announcement follows several model and access changes across Google Search:

- Gemini 4 Hydrogen is the general AI Mode default.[1][3]
- Gemini 4 Hydrogen is routed to the hardest AI Overview questions.[1]
- Gemini 4 Argon is available as an additional model for Google AI Pro and Ultra subscribers through AI Mode.[4][5]
- Gemini 4 Carbon is available in AI Mode to Google AI Ultra subscribers.[6]
- Gemini 2.5 Pro remains the model explicitly named for Deep Search in Google’s help documentation.[7][8]

The clearest answer to “which Gemini model powers AI Mode?” is therefore **Gemini 4 Hydrogen**. The answer to “which Gemini model powers Google Search everywhere?” is more complicated because Google describes different defaults, routes, subscriber options and specialised workflows.[9]

## Why does Gemini 4 Hydrogen matter?

Google says Hydrogen changes how AI Mode understands questions, searches for information and presents answers. The company describes improvements in complicated-intent understanding, instruction following, query fan-out, links, visual responses and language nuance.[10]

Google’s claims describe product capabilities rather than an independently measured accuracy result. The announcement does not provide a quantified before-and-after Search accuracy improvement, so Hydrogen should not be described as guaranteeing more accurate answers.[1]

### Reasoning and complicated questions

Google says Hydrogen can better handle multi-part questions, typos and specific constraints. In practical terms, that means the model is intended to interpret what a user wants before deciding how to answer.[10][7]

This matters because AI Mode is designed for questions that may require several steps rather than a single conventional search. A user may ask for a comparison, impose several conditions and expect the answer to use current information. Google says Hydrogen is better at following that type of instruction.[10]

### Query fan-out and retrieval

Google says improved reasoning strengthens query fan-out and helps connect users with more useful links.[10] Query fan-out is the process of generating multiple related searches across different subtopics while answering one original question.[5][11]

For example, Google’s documentation uses a lawn-repair question to illustrate related searches about chemical herbicides, non-chemical weed removal and future prevention. The point is not to repeat the original query several times, but to investigate different aspects of the user’s goal.[12]

Google says AI Mode divides a question into subtopics, searches them concurrently across multiple data sources and combines the results into a response.[13][11] Hydrogen therefore matters not only because it writes the final answer, but also because Google says its reasoning improves how Search decides what related information to retrieve.[14][13]

### Visual and interactive answers

Google says Hydrogen can produce more images, structured responses, comparison tables and custom tools.[1][10] This extends AI Mode beyond a plain text summary and makes the format of the response part of the model upgrade.

Google previously used an interactive three-body-problem simulation to demonstrate Gemini 3 in Search. Users could manipulate variables and observe gravitational interactions while the system explained the subject.[14] That example showed how a Search model can affect both the information retrieved and the way an answer is presented.[13][15]

Hydrogen’s announcement continues that direction with adaptive visual responses, tables and tools, but it does not establish that every question produces an interactive result.[10][1]

## What is query fan-out in Google AI Mode?

Query fan-out is Google’s term for generating several related searches to answer a broader question. AI Mode uses those searches to investigate subtopics concurrently, retrieve additional information and assemble a response with supporting links.[13][11]

Fan-out is different from retrieval-augmented generation, or RAG. Google describes RAG, also called grounding, as retrieving relevant and up-to-date pages from the Search index and using information from those pages to generate a response with links.[12]

The distinction is useful:

1. **Interpretation:** the system identifies the user’s goal and constraints.[16][5]
2. **Fan-out:** the system expands the request into related sub-queries.[13]
3. **Retrieval:** Search finds relevant pages and other information.[12]
4. **Generation:** the model produces a response grounded in the retrieved material.[12]

Google’s Gemini 3 announcement said improved reasoning could perform more searches and find relevant material that earlier approaches might miss.[14][10] The Hydrogen announcement similarly connects reasoning improvements with broader fan-out and more helpful links.[10]

Fan-out is not a guarantee that Search finds every relevant source. Google warns that AI Mode can misinterpret web content or miss context, and recommends checking important information in more than one place.[5]

### How fan-out relates to RAG

Academic research provides background for the retrieval-and-generation pattern. Lewis and colleagues’ NeurIPS 2020 paper studied retrieval-augmented generation systems that combine a model’s learned memory with information retrieved from an external index.[17]

The researchers reported more specific, diverse and factual generated language than a parametric-only baseline in their research setting.[17] That paper does not establish that Google AI Mode uses the same architecture, and its results should not be presented as a measurement of Hydrogen’s performance.[15][11]

For Google Search, the safe explanation is that fan-out broadens the searches performed for a question, while grounding connects the generated response to retrieved information.[12]

## What does Deep Search use?

Google’s dedicated Deep Search help page explicitly identifies **Gemini 2.5 Pro** as the model behind Deep Search in AI Mode.[18][14] The page says Deep Search browses hundreds of sites, reasons across the results and creates a comprehensive cited report.[7]

This is a separate disclosure from the November Hydrogen announcement. Google has not published a cited statement in the supplied documentation saying that Deep Search has moved from Gemini 2.5 Pro to Hydrogen, Argon or Carbon.[6]

Deep Search follows a distinct workflow:

- Users open AI Mode and select the Deep Search icon before submitting a question.[7]
- Google may ask clarifying questions before beginning the research.[7]
- The system browses hundreds of sites and creates a cited report.[7]
- Reports usually take about five minutes, although complex requests may take longer.[7]
- Users can begin another search while the report is being prepared and revisit past reports later.[7]

Google documents Deep Search for US English users in Search Labs on desktop and mobile web, with access for Google AI Pro and AI Ultra subscribers.[7] The documentation also says users must be at least 18 and signed in to a personally managed Google Account.[5][7]

Deep Search should therefore be described as a research workflow with its own documented model, not as a synonym for every long response generated by AI Mode.

## What is confirmed, and what is not?

The available Google statements support several precise conclusions.

### Confirmed

- Gemini 4 Hydrogen is the new default model in AI Mode as of Google’s 12 November 2026 announcement.[1]
- Hydrogen is available globally in Search for all users.[10][1]
- Google uses Hydrogen for the hardest questions in AI Overviews.[1]
- Gemini 4 Argon is available to Google AI Pro and Ultra subscribers through AI Mode, subject to the documented limits.[5][6]
- Gemini 4 Carbon is available in AI Mode to Google AI Ultra subscribers.[6]
- Google’s Deep Search help page names Gemini 2.5 Pro.[7][8]
- Google describes query fan-out as concurrent related searches across subtopics.[13][11]

### Not confirmed

Google has not published the complete model routing for all AI Overview questions in the November announcement.[1] It is therefore not confirmed that every AI Overview uses Hydrogen.

It is also not confirmed in the cited Deep Search documentation that Hydrogen has replaced Gemini 2.5 Pro for that workflow.[1] The latest explicit model disclosure for Deep Search remains Gemini 2.5 Pro.[15]

The existence of a newer default also does not prove that every specialised Search feature uses the same model. AI Mode defaults, AI Overview routing, subscriber model selection and Deep Search are separate product statements.[9]

## How did Google Search’s Gemini models change over time?

Google’s model history shows why a single phrase such as “Gemini-powered Search” is no longer sufficiently precise.

| Date | Model or model family | Search surface | What Google established |
|---|---|---|---|
| 5 March 2025 | Gemini 2.0 and a custom Gemini 2.0 | AI Overviews and experimental AI Mode | Gemini 2.0 launched for AI Overviews in the US, while experimental AI Mode used a custom version for an initial Labs audience.[13] |
| 20 May 2025 | Custom Gemini 2.5 | AI Mode and AI Overviews | Google brought a custom Gemini 2.5 to both surfaces in the US.[19][15] |
| 16 July 2025 | Gemini 2.5 Pro | AI Mode and Deep Search | Gemini 2.5 Pro became selectable in AI Mode for an initial subscriber rollout and was named as the model behind Deep Search.[8] |
| 18 November 2025 | Gemini 3 Pro | AI Mode and planned routing | US AI Pro and Ultra subscribers could select Thinking in AI Mode; Google also announced planned routing of difficult questions to Gemini 3.[14][16] |
| 17 December 2025 | Gemini 3 Flash | AI Mode | Gemini 3 Flash became the global AI Mode default.[2] |
| 27 January 2026 | Gemini 3 | AI Overviews | Gemini 3 became the global default for AI Overviews.[20][1] |
| 19 May 2026 | Gemini 3.5 Flash | AI Mode | Gemini 3.5 Flash became the new AI Mode default globally.[21] |
| 10 November 2026 | Gemini 4 Carbon | AI Mode | Carbon became available in AI Mode to Google AI Ultra subscribers.[6] |
| 12 November 2026 | Gemini 4 Hydrogen | AI Mode and difficult AI Overview questions | Hydrogen became the new AI Mode default and was identified for the hardest AI Overview questions.[1] |

The timeline separates three ideas that are often conflated: the model Google selects by default, models subscribers can choose, and models used for specific features.[1][17]

The May and July 2025 announcements are an early example. May described a custom Gemini 2.5 for AI Mode and AI Overviews, while July specifically named Gemini 2.5 Pro for selectable AI Mode access and Deep Search.[15][7] Those announcements did not describe identical roles.

## What should publishers and search marketers take from the update?

The Hydrogen update makes the answer-generation layer more capable, but Google’s own documentation continues to describe Search as a retrieval-and-generation system. A model interprets the question, fan-out expands the research, Search retrieves information and the system generates a linked response.[5][13]

For publishers, the practical implication is not that one model name guarantees visibility. Google’s documentation says AI Mode searches multiple related subtopics and uses retrieved pages to support its response.[12] Content that clearly addresses distinct aspects of a question may therefore be easier for the system to use, but Google does not provide a guarantee of inclusion or citation.

For users, newer reasoning does not remove the need to verify important information. Google warns that AI Mode can make mistakes, misinterpret web content or miss context.[5] Deep Search produces a fully cited report, but citations still need to be reviewed for relevance and accuracy.[15]

The most accurate November 2026 summary is:

> Google AI Mode defaults to Gemini 4 Hydrogen; Hydrogen also handles the hardest AI Overview questions, while Deep Search documentation still names Gemini 2.5 Pro.[1]


Background: what are AI Mode and AI Overviews?

AI Overviews and AI Mode are separate Google Search experiences. Google’s 2025 documentation describes AI Overviews as search-result summaries and AI Mode as a more conversational experience designed for follow-up questions and broader exploration.[20][13]

AI Mode can break a question into related searches through query fan-out, retrieve information from multiple sources and generate a response with supporting links.[5][12] AI Overviews can also use Gemini models, but the model assignment can vary by announcement, question difficulty and rollout stage.[14][13]

That is why “Gemini powers Google Search” is an incomplete description in November 2026. The specific answer depends on the Search surface, date, default status, subscriber access and feature being discussed.[8][1]



## FAQ
### Which Gemini model powers Google AI Mode in November 2026?
Gemini 4 Hydrogen is the new default model in Google AI Mode, according to Google’s 12 November 2026 announcement.[1] Google says Hydrogen is available globally in Search for all users.[1]

### Does every Google AI Overview use Gemini 4 Hydrogen?
Google has not said that every AI Overview uses Hydrogen. The November announcement says Hydrogen is used for users’ hardest AI Overview questions, leaving the model allocation for other questions unpublished.[1]

### What is Gemini 4 Hydrogen changing in AI Mode?
Google says Hydrogen improves the handling of complicated questions and instructions, query fan-out, links, visual responses and language nuance.[1][10] These are Google’s stated product capabilities, not a published independent measurement proving a universal accuracy increase.[15]

### What does query fan-out mean in Google Search?
Query fan-out means generating multiple related searches across subtopics to answer one broader question.[10][15] Google says AI Mode performs these searches concurrently, retrieves information and brings the results together in a response.[13][11]

### Does Deep Search use Gemini 4 Hydrogen?
Google’s dedicated Deep Search help page explicitly names Gemini 2.5 Pro.[7][8] The cited documentation does not confirm that Deep Search has migrated to Hydrogen or another Gemini 4 model.[9]

## Sources
[1] Google Makes Gemini 4 Hydrogen The Default Model In AI Mode — https://www.searchenginejournal.com/google-gemini-4-hydrogen-default-ai-mode/597096/
[2] Introducing Gemini 3 Flash: Benchmarks, global availability — https://deepmind.google/blog/gemini-3-flash-frontier-intelligence-built-for-speed/
[3] Robby Stein on X: "(1/6) The biggest model upgrade to Search yet has landed. Gemini 4 Hydrogen is here, and brings cutting-edge intelligence to all Search users. It’s now the new default model in AI Mode & being used for your hardest questions in AI Overviews.

> Breaks down complicated, multi-part… / X — https://x.com/rmstein/status/2120941443359576166
[4] Gemini 4 Argon is now available for everyone — https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon?hl=en
[5] Get AI-powered responses with AI Mode in Google Search - Computer - Google Search Help — https://support.google.com/websearch/answer/16011537?co=GENIE.platform%3DDesktop&hl=en
[6] Gemini 4 Carbon is now available — https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/?utm_source=deepmind.google&utm_medium=referral&utm_campaign=gdm&utm_content=
[7] Use Deep Search in AI Mode for in-depth research - Google Search Help — https://support.google.com/websearch/answer/16420827?hl=en
[8] New AI features in Google Search: Call a business or do research — https://blog.google/products-and-platforms/products/search/deep-search-business-calling-google-search/
[9] Introducing Gemini 4 Hydrogen for Agentic Workflows — https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen
[10] Google Search upgrade: Gemini 4 Hydrogen features — https://blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade
[11] AI Overviews and AI Mode in Search — https://search.google/pdf/google-about-AI-overviews-AI-Mode.pdf
[12] Google's Guide to Optimizing for Generative AI Features on Google Search | Google Search Central | Documentation | Google for Developers — https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
[13] Expanding AI Overviews and introducing AI Mode — https://blog.google/products-and-platforms/products/search/ai-mode-search/
[14] Google brings Gemini 3 AI model to Search and AI Mode — http://blog.google/products-and-platforms/products/search/gemini-3-search-ai-mode
[15] AI Mode in Google Search: Updates from Google I/O 2025 — https://blog.google/products/search/google-search-ai-mode-update
[16] AI Mode update: Gemini 3 Flash, Nano Banana Pro — https://blog.google/products-and-platforms/products/search/google-ai-mode-update-gemini-3-flash
[17] Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks — https://proceedings.nips.cc/paper/2020/hash/6b493230205f780e1bc26945df7481e5-Abstract.html
[18] Gemini 3: Introducing the latest Gemini AI model from Google — https://deepmind.google/blog/a-new-era-of-intelligence-with-gemini-3
[19] AI Overviews expand to over 200 countries and territories, more than 40 languages — https://blog.google/products-and-platforms/products/search/ai-overview-expansion-may-2025-update/
[20] AI Mode in Google Search and AI Overviews get Gemini upgrades — https://blog.google/products-and-platforms/products/search/ai-mode-ai-overviews-updates/?trk=article-ssr-frontend-pulse_little-text-block
[21] Google Search’s I/O 2026 updates: AI agents and more — https://blog.google/products-and-platforms/products/search/search-io-2026
