Market Signal · AI and LLMs

Google's new default AI Mode model, Gemini 4 Hydrogen, adds more links and comparison tables

Gemini 4 Hydrogen now answers AI Mode questions for every Google user and handles the hardest AI Overviews. Google promises more links and more visual answers, but gives no counts.

By Market Signal desk, WebVisiblityPublished Updated 7 min read
Dot chart comparing Gemini 4 Hydrogen and Claude Sonnet 5.5 on nine tests from Google's model card. Hydrogen leads on Agents' Last Exam v1.1 (71.6% vs 60.4%), SWE-sweep (13.5% vs 3.1%), HLE-Diamond (58.8% vs 53.3%), Chartography (66.6% vs 61.6%), AutomationBench-AA (76.1% vs 71.8%) and MKA (53.4% vs 52.0%). Sonnet 5.5 leads on OSWorld-2.1 (83.9% vs 77.6%), Terminal-Bench 4.1 (63.9% vs 57.0%) and Terminal-Bench-Science 0.1 (59.9% vs 51.4%).
Google's model card lists 10 tests. Sonnet 5.5 leads on 3 (counted from the card); Hydrogen leads on the other 7, including Vending-Bench 2 ($14,549 vs $11,165), which is left out of the chart because it is scored in dollars. Chart: WebVisiblity. Data: Google DeepMind, Gemini 4 Hydrogen model card

Google switched the model behind AI Mode on 12 November. Gemini 4 Hydrogen is rolling out globally in Google Search for all users, and Google's Robby Stein said it is now the new default model in AI Mode and is used for the hardest questions in AI Overviews. Google calls it the biggest model upgrade to Search yet.

At a glance

ModelGemini 4 Hydrogen, adapted from Gemini 4 Carbon
Launched in Search12 November 2026, for all users worldwide
Role in AI ModeThe new default model
Role in AI OverviewsUsed for your hardest questions, per Google's Robby Stein
API price$1.50 in, $7.50 out per million tokens
Context windowUp to 1 million tokens
Knowledge cutoffMarch 2026

The signal

  • Hydrogen is now the default model in AI Mode for users worldwide, Search Engine Journal reports.
  • Google says the upgrade brings a broader range of helpful links through improved query fan-out.
  • Answers get more images, structured responses, comparison tables and custom tools.
  • Google's own model card shows Claude Sonnet 5.5 ahead on three of the tests Google published.
  • Google gave no link or site counts, so the extra-links claim cannot yet be measured.
  • The switch landed while the November core update was still rolling out.

What happened

On Thursday 12 November, Google released Gemini 4 Hydrogen across its products and put it into Search the same day. The Search post, written by Robby Stein, VP of Product for Google Search, says the model is available globally in Google Search for all users. Stein wrote on X that it is the new default model in AI Mode and that it also handles the hardest questions in AI Overviews.

Google lists four changes for searchers. The model reads intent better, including multi-part questions, typos and very specific requests. Improved query fan-out connects people to a broader range of helpful links. Answers become more visual, with more images, structured responses, comparison tables and custom tools. Language understanding improves across all supported AI Mode languages. Rajan Patel of Google described the link change as even more links to help people find sites across the web.

Hydrogen is the speed tier of Google's new family. Google's model post says it is built for rapid execution across diverse agentic workflows, while Gemini 4 Carbon and Gemini 4 Argon handle deep, long-horizon reasoning. The model card says Hydrogen is adapted from Gemini 4 Carbon. Search Engine Journal found the AI Mode menu now labels the models Everyday help (4 Hydrogen) and Most complex work (4 Carbon), with Carbon limited to Ultra subscribers.

The same model is also in the Gemini app, Google Sheets and Google Antigravity, Thurrott reports, with Antigravity access for AI Pro and Ultra subscribers. Developers pay $1.50 per million input tokens and $7.50 per million output tokens across the full context window.

Bar chart of Artificial Analysis Intelligence Index scores reported by Google: Gemini 4 Hydrogen at xhigh thinking 54, Hydrogen at its default high setting 51, Gemini 4 Argon 50.
Scores as reported in Google's launch post; no calculation added. High is Hydrogen's default thinking level. Chart: WebVisiblity. Data: Google, Introducing Gemini 4 Hydrogen

By the numbers

Gemini 4 Hydrogen by Google's own numbers
 51Artificial Analysis Intelligence Index, Hydrogen at defaultGoogle
 57%Terminal-Bench 4.1, HydrogenGoogle
 $1.50API input price per million tokensGoogle
 $7.50API output price per million tokensGoogle
 3 of 10Google tests where Sonnet 5.5 leadsGoogle DeepMind

Timeline

  1. Gemini 4 Argon becomes an option for AI Pro and Ultra subscribers in AI Mode.

  2. Gemini 4 Carbon starts rolling out to Ultra subscribers in AI Mode; the November core update begins the same day.

  3. Google releases Gemini 4 Hydrogen and makes it the new default model in AI Mode.

“It's the biggest model improvement to Search yet. And it's available now for all users, globally.”

Nick Fox, Google, on X · source

What Google says changes in AI Mode answers

  1. 1Reads intent

    Handles multi-part questions, typos and specific requests.

  2. 2Fans out wider

    Links to a broader range of helpful links.

  3. 3Shows more

    Adds images, structured responses and comparison tables.

  4. 4Understands languages better

    Better understanding across all supported AI Mode languages.

Why it matters

For B2B brands, the point is not the benchmark race but where the answer sends people. Google says Hydrogen widens query fan-out to reach a broader range of helpful links. If AI Mode now cites more sources per answer, mid-sized vendors that were left out of short answers have more room to be named. Google has not said how many more links appear, and Search Engine Journal notes the posts give no link or site counts, so this has to be checked query by query.

The second change is format. Google says answers will carry more comparison tables and custom tools, and Android Central's write-up of the update points to images, charts and graphs, and sometimes tables. Vendor shortlists are the kind of question that suits a table. A table needs clean, current facts such as plans, prices, integrations and limits, so the brands whose pages state those plainly are the easiest to place in one.

The third point is caution about Google's own claims. Its model post says Hydrogen cuts fabricated claims by more than 5x compared with Gemini 3.8 Flash, which matters to any brand that has seen AI answers misstate its pricing. Yet the model card also shows Claude Sonnet 5.5 ahead on OSWorld-2.1 at 83.9% against 77.6% and on both Terminal-Bench tests. The model is strong but not best at everything, so measure what it says about your brand rather than trust the launch post.

Who is affected

High impactB2B vendors outside the usual top results

Wider fan-out could cite more sites, though Google gave no link counts.

High impactProduct and pricing page owners

More comparison tables reward clear, current specs.

Medium impactSEO and analytics teams

The core update overlaps the model switch, which muddies traffic reads.

Low impactDevelopers building on Gemini

A flat $1.50 and $7.50 per million tokens across the full context.

What to do now

For marketers

  • Run your 20 most important buyer questions in AI Mode this week and save each answer with its cited links.
  • Note which answers now show a comparison table and whether your brand is a row in it.
  • Put plans, prices, integrations and limits in plain text on one page so a table can quote them.
  • Mark the week of 10 November in your reports so the core update and the model switch are not confused.

For software buyers and IT teams

  • Treat AI Mode vendor tables as a starting list and check each figure on the vendor's own site.
  • Ask the same question in AI Mode and one other assistant, since Google's own tests show no single model wins everything.
Checklist: test your brand in the new AI Mode

Printable, one page, 10 checks. Enter your email and it opens right away.

We email you a copy. We never share your email and never send you sales emails.

These steps are our recommendations based on this news, not statements from the companies involved.

What to watch

  • Link counts. Google promised more links but no figures; independent trackers will be the first to show how many.
  • AI Overviews. Only the hardest questions use Hydrogen in AI Overviews; watch whether Google widens that.
  • Mixed signals. The November core update overlaps the switch, so separate the two before blaming either for traffic changes.
  • Carbon for more users. Today Carbon is for Ultra subscribers with daily limits; a wider release would change complex B2B research queries again.
  • Safety report. Google says the full safety and security report is coming soon.
Bar chart of Terminal-Bench 4.1 scores published by Google: Gemini 3.8 Flash 18.2%, Gemini 4 Hydrogen 57%, a gap of 3.1 times.
3.1x is 57 divided by 18.2 (3.13), rounded. Both scores are Google's own figures. Chart: WebVisiblity. Data: Google, Introducing Gemini 4 Hydrogen

Questions readers ask

What is Gemini 4 Hydrogen?

Google's fast model for agentic work, released on 12 November 2026 and adapted from Gemini 4 Carbon. It now powers AI Mode in Google Search for all users.

Is Gemini 4 Hydrogen the default model in Google AI Mode?

Yes. Google's Robby Stein said Hydrogen is the new default model in AI Mode, worldwide, from 12 November 2026.

Does Gemini 4 Hydrogen change AI Overviews?

Partly. Google's Robby Stein said Hydrogen is used for your hardest questions in AI Overviews. Google has not said what share of AI Overviews that covers.

Will Gemini 4 Hydrogen show more website links in AI Mode?

Google says it connects people to a broader range of helpful links, but it has published no counts, so the change must be measured query by query.

How much does the Gemini 4 Hydrogen API cost?

$1.50 per million input tokens and $7.50 per million output tokens, flat across its 1 million token context window.

Is Gemini 4 Hydrogen better than Claude Sonnet 5.5?

On some tests. Google's model card shows Hydrogen ahead on most of its 10 tests, but Claude Sonnet 5.5 leads on OSWorld-2.1 and both Terminal-Bench tests.

What should B2B marketers do about Gemini 4 Hydrogen in AI Mode?

Re-run your key buyer questions in AI Mode now. Check which sources are cited and whether answers show comparison tables that include or skip your brand.

Companies in this story

Sources

  1. GoogleOriginal
  2. Google
  3. Google DeepMind
  4. Search Engine Journal
  5. Search Engine Roundtable
  6. Thurrott
  7. Android Central

Researched, fact-checked and edited by the Market Signal desk. Read our editorial standards. Spotted an error? Email hello@webvisiblity.com and we will correct it.

Related briefings

AI and LLMsOct 29, 2026, 6:30 AM ET8 min read

Google Search growth slowed to 12% as AI Mode earned less per search

Alphabet's third quarter shows record query growth but the slowest Search revenue growth this year, and Google named lower AI Mode monetization as one cause. It is now putting ads into AI Mode follow-up questions. Here is what changed for B2B marketers.

AI and LLMsNov 10, 2026, 5:30 AM ET8 min read

Claude Fable 5.2 is 20% cheaper, yet Anthropic still tells most buyers to start with Opus

Anthropic's most capable public model now costs $8 per million input tokens and $40 per million output tokens. That is still twice the price of Opus 5.5, and Anthropic's own benchmark table shows rivals winning some work tasks.

AI and LLMsOct 31, 2026, 1:04 AM ET8 min read

Bing will let sites block AI training but keep AI answers, as rivals back one robots.txt token

A single robots.txt group now carries a site's AI training preference to several AI companies. Microsoft went further and split training from grounding, the step where a page feeds a live AI answer. For B2B brands, those two switches now have very different costs.

AI and LLMsOct 27, 2026, 12:46 PM ET9 min read

Anthropic files to go public, and its IPO filing shows how Claude grew 14 times in a year

The S-1 behind the ANT ticker reveals $16.3 billion of revenue in six months, a headline loss that is mostly paper, and the playbook that turned a coding tool into a workplace habit. Here is what it means for B2B marketers.

AI Mode just changed its model. Is your brand still in the answer?

The free check shows what ChatGPT, Gemini, Claude and Perplexity say about your brand in about 2 minutes.