Google has put a new model behind its search engine. Gemini 4 Hydrogen, released on 12 November 2026, is now the default model in Google AI Mode and handles the hardest questions in AI Overviews, while wider Search receives what Google calls its biggest model upgrade yet. It is live globally for all users, and it arrived two days into the November 2026 core update, which is still rolling out. For site owners, the question is not whether Hydrogen rewrites the ranking rules — Google has said nothing of the kind — but how a faster, more visual and more link-hungry answer layer changes what searchers see before they ever reach a website.
Key takeaways
- Gemini 4 Hydrogen launched on 12 November 2026 and is now the default model in Google AI Mode, used for the hardest AI Overview questions.
- Google calls it the biggest model upgrade to Search yet: better intent understanding, wider query fan-out, more visual answers and richer language understanding.
- Hydrogen is the fast execution model of the Gemini 4 family; Carbon and Argon handle deep, long-horizon reasoning.
- Google reports tasks 2.5 times faster than Gemini 3.8 Flash, but every performance figure is Google’s own, not an independent test.
- Do not read traffic changes into Hydrogen yet: the November core update is still rolling out, and a logging error undercounted Generative AI impressions for 9 to 11 November.
What Google announced on 12 November
Two announcements landed on the same day. Robby Stein, Google’s Vice President of Product for Search, wrote on the Google Keyword blog that Hydrogen brings cutting-edge intelligence to everyone and called it the biggest model upgrade to Search yet, available globally for all users. Koray Kavukcuoglu of Google DeepMind introduced the model itself as the speed and efficiency member of the Gemini 4 family, built for everyday coding and agentic workflows. Anthropic launched Claude Fable 5.2 three days earlier.
Carbon and Argon are the frontier models for deep, long-horizon reasoning; Hydrogen is built for rapid execution: writing code, running tests, verifying changes, and carrying out multi-step tasks. Gemini 4 reaches every tier at once: the Gemini API, AI Studio, Gemini for Business, the Gemini app, Google Search, and Antigravity for paid Pro and Ultra subscribers.
Hydrogen offers four thinking levels — low, medium, high and a new xhigh tier — with high as the default. Its context window is 1 million tokens, priced flat in the API at $1.50 per million input tokens and $7.50 per million output tokens. Using it inside Search costs searchers nothing extra.
What actually changes inside Search
Stein’s post lists four changes.
First, better understanding of intent: multi-part questions, hard-to-read typos and highly specific requests, with closer instruction following and more precise tool calling, so each part of a question gets an answer.
Second, improved query fan-out and linking: simultaneous queries across the subtopics of a question, connecting searchers to what Google calls a broader range of helpful links than before, including sites they might not have discovered. Google gives no counts, so the claim can only be checked against your own citation and referral data.
Third, more visual responses: more images, structured responses, comparison tables and custom tools, with answers that adapt to the searcher and stay easy to scan.
Fourth, richer language understanding across all supported AI Mode languages, so answers stay nuanced in any language.
The fan-out change is the one site owners should watch most closely. A single messy question becomes several parallel sub-queries, and the pages that answer a subtopic clearly stand the best chance of being cited.
The AI Mode default and the hardest AI Overview questions
The two most consequential details did not appear in the Keyword blog post. They came from Stein’s post on X, as reported by Search Engine Journal: Hydrogen is now the default model in AI Mode, and it is being used for the hardest questions in AI Overviews.
That replaces Gemini 3.5 Flash, the global AI Mode default since Google I/O in May 2026. In AI Overviews the picture is more selective: Google made Gemini 3 the default globally in January 2026, and Hydrogen now takes only the hardest questions. Google has not said which questions count as hardest, or what answers the rest. The AI Mode help page describes Hydrogen as helpful for fast, all-around assistance on most questions.
The practical consequence is a date to log: citation data for AI Mode or AI Overviews this month now spans a model change on 12 November, and any comparison that ignores that date mixes two different answer engines.
How Gemini 4 Hydrogen measures up
Google’s DeepMind announcement carries the headline figures. Every one of them is Google’s own, with no independent replication yet.
| Claim | Hydrogen | Comparison |
|---|---|---|
| Task speed at the high default | 2.5 times faster | Gemini 3.8 Flash |
| Terminal-Bench 4.1 | 57 percent | 18.2 percent for 3.8 Flash |
| Fabricated claims | Cut by more than 5 times | 3.8 Flash |
| Artificial Analysis Intelligence Index | 51 at high, 54 at xhigh | 50 for Argon |
| Partner testing | 2x to 2.8x faster, 12 to 23 percent fewer tokens | 3.8 Flash |
The safety figures follow the same pattern. Google says Hydrogen cuts overreach in agentic tasks by up to 81 percent and under-service by up to 79 percent, reduces misaligned actions and task shortcutting by 12 times, and reaches a 96.7 percent honesty rate on its internal evaluation. These are company-reported results, and they should be read that way until independent testing exists.
What Hydrogen does not change
Google has not said Hydrogen changes regular rankings. The Keyword blog post says nothing about ranking, and no new guidance for site owners came with the launch. The standing position is that AI Overviews and AI Mode need nothing beyond normal search optimisation.
The November 2026 core update began on 10 November and may take up to two weeks. As of 16 November there is still no completion entry on the Search Status Dashboard. Any traffic movement this month is tangled with that rollout, and Hydrogen landed in the middle of it. Do not attribute a traffic change to Hydrogen without separating the two dates.
The Search Console trap: ignore the 9 to 11 November dip
There is a second date to mark, and it is a trap. Google’s Search Console data anomalies page records a logging error that cut impressions on the Generative AI performance report in Search for data from 9 to 11 November 2026, adding that the issue affects data logging only.
That window covers the day the core update began, immediately before Hydrogen’s launch. A dip in that report on those three days is a counting fault, not a sign that Google stopped showing a site, and Google has not said whether the missing impressions will be restored. Leave those three days out of any before-and-after comparison.
What site owners should do now
- Wait for the core update to finish: confirm completion on the Search Status Dashboard, then wait at least a full week before judging the effect. That has not happened yet.
- Annotate three dates: 9 to 11 November for the logging error, 10 November for the core update, and 12 November for Hydrogen.
- Check AI Mode answers for your key queries: which pages get cited, and do your pages answer the subtopics fan-out creates?
- Keep pages citable: direct answers near the top, clear headings, and tables where data helps are the structures a fan-out answer can lift.
- Watch your own numbers after Gemini 4 Hydrogen: citation and referral data after 12 November, compared with before, is the evidence that matters for your site.
- Is Gemini 4 Hydrogen now the default model in Google Search?
- Does Gemini 4 Hydrogen change Google rankings?
- How is Hydrogen different from Carbon and Argon?
- How much does Gemini 4 Hydrogen cost?
- Why did my Generative AI impressions fall on 9 to 11 November?
Frequently asked questions
Is Gemini 4 Hydrogen now the default model in Google Search?
In AI Mode, yes, for all users worldwide. In AI Overviews it handles the hardest questions, and Google has not defined which those are. The wider Search experience uses Hydrogen globally.
Does Gemini 4 Hydrogen change Google rankings?
Google has not said so. The announcement describes changes to how Search understands questions, finds links and presents answers. Ranking changes come from core updates, and the November 2026 core update is still under way.
How is Hydrogen different from Carbon and Argon?
Hydrogen is the fast execution model, built for everyday coding, agentic workflows and rapid multi-step tasks. Carbon and Argon are the frontier models for deep, long-horizon reasoning. In Antigravity, Google’s Orchestrator switches between the two inside a session.
How much does Gemini 4 Hydrogen cost?
In the Gemini API it costs $1.50 per million input tokens and $7.50 per million output tokens, flat across its 1 million-token context window. Inside Search and the Gemini app there is no charge.
Why did my Generative AI impressions fall on 9 to 11 November?
Because of a logging error, not a visibility loss. Google’s data anomalies page says a logging error cut impressions on the Generative AI performance report in Search for those three days; the issue affects data logging only.
Sources and Further Reading
- Google Keyword blog, Robby Stein, “Google Search gets a major upgrade with Gemini 4 Hydrogen”, 12 November 2026
- Google DeepMind blog, Koray Kavukcuoglu, “Introducing Gemini 4 Hydrogen for Agentic Workflows”, 12 November 2026
- Search Engine Journal, “Google Makes Gemini 4 Hydrogen The Default Model In AI Mode”, 12 November 2026
- Google Search Status Dashboard, November 2026 core update incident
- SEO Chat, “Core update week: read your Google traffic the right way”, 16 November 2026
