What Is Gemini 4 Hydrogen and How to Use It

By Omkar Tilekar · Updated 16 November 2026 · 5 min read

Quick answer: Gemini 4 Hydrogen is Google's newest AI model, launched on 12 November 2026 and built for agentic work — long, multi-step jobs like coding, research and turning dense documents into cited reports. It costs $1.50 per million input tokens and $7.50 per million output tokens, and it is 2.5× faster than Gemini 3.8 Flash at its default setting. Best of all, you do not need to pay or sign up for anything new to try it: it is already rolling out to every Google Search user and to free Gemini app users through automatic model selection.

KEY FACTS — GEMINI 4 HYDROGEN

✓ Launched 12 Nov 2026 — Google's newest model  |  ✓ $1.50 / $7.50 per million tokens  |  ✓ Free to try — already in Google Search and the Gemini app

Key highlights

  • Gemini 4 Hydrogen launched 12 November 2026, aimed at agentic workflows: coding, research, and long tasks that need many steps and tool calls.
  • Pricing is flat across the full 1-million-token context window: $1.50 per million input tokens, $7.50 per million output tokens.
  • It scores 51 on the Artificial Analysis Intelligence Index at default "high" thinking (vs 50 for Gemini 4 Argon), and 54 at the new "xhigh" level.
  • On Terminal-Bench 4.1 it scores 57%, against 18.2% for Gemini 3.8 Flash — and it runs 2.5× faster than Flash at the default setting.
  • Four thinking levels — low, medium, high (default) and xhigh — so you can trade speed for depth on hard tasks.
  • Free users get it automatically in Search and the Gemini app; developers get it in the Gemini API and Google AI Studio (all tiers).

In this guide

Google launches a new model every few months, so the fair question is: why does this one matter? Because Hydrogen is not a slightly better chatbot. It is the model Google built for work that takes many steps — the kind of job where an AI has to read a long document, use tools, check its own work and come back with something finished. And unlike most launches, this one reaches ordinary users on day one: it is already in Google Search for everyone, not locked behind a paid tier.

What is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is the newest member of Google's Gemini 4 family, announced on 12 November 2026 in the official Google blog post "Introducing Gemini 4 Hydrogen for Agentic Workflows" by Koray Kavukcuoglu, SVP at Google DeepMind. "Agentic" means the model is designed to act: it can plan a multi-step task, call tools, work through a long document or codebase, and keep going until the job is done — rather than answering one question and stopping.

Hydrogen sits alongside two siblings you may already know. Gemini 4 Carbon is the most capable model in the family — the one for the hardest reasoning. Gemini 4 Argon is the balanced model. Hydrogen is the fast, efficient one: close to Argon's quality on the standard benchmarks, but much quicker and cheaper to run. Google's own framing is that Hydrogen is the upgrade path from Gemini 3.8 Flash — the previous fast model — not a replacement announced with a shutdown date. There is no Flash switch-off date, so do not believe any guide that claims one.

It is rolling out from 12 November to the Gemini API and Google AI Studio (all tiers), the Gemini app, Google Search (all users), Gemini for business, and Google Antigravity for paid Pro and Ultra subscribers. In the Gemini app, free users get it through automatic model selection — you do not pick it from a menu, the app routes suitable tasks to it. In Google Antigravity, a new Orchestrator switches between Carbon and Hydrogen inside one session: on Terminal-Bench 4.0 it reached near-Carbon success while using 24% less quota and finishing 19% faster.

How fast and how smart is it?

The numbers that matter, all from Google's launch post:

BenchmarkGemini 4 HydrogenWhat it means
Artificial Analysis Intelligence Index (default "high" thinking)51One point ahead of Gemini 4 Argon (50) — near-top quality at a fast model's price
Intelligence Index at "xhigh" thinking54The new top thinking level pushes Hydrogen past Argon when you need maximum depth
Terminal-Bench 4.1 (agentic coding tasks)57%Against 18.2% for Gemini 3.8 Flash — a three-fold jump on real multi-step coding work
Speed vs Gemini 3.8 Flash (default setting)2.5× fasterThe everyday feel: answers arrive noticeably quicker for the same kind of task
Context window1 million tokensAbout 750,000 words — whole books, long reports, large codebases in one go
Price$1.50 / $7.50 per million input / output tokensFlat across the full context window — no price jump for long inputs

Hydrogen also introduces four thinking levels — low, medium, high (the default) and xhigh. Thinking levels are how hard the model works before it answers: low is quick and cheap for simple questions, high is the balanced default, and xhigh spends more compute on the hardest problems. The practical rule is simple: leave it on high for everyday work, and switch to xhigh only when a task is genuinely hard — a long research job, a tricky piece of code, a document you need checked carefully.

Feeding Hydrogen a big PDF? Shrink it first, free →
Hydrogen is built to synthesise dense PDFs into cited reports — but a 40 MB report uploads slowly and wastes tokens. MeraToolkit's free PDF Compressor shrinks a PDF towards a 100, 200 or 500 KB target in your browser, no sign-up, and your file never leaves your device.
Open the free PDF Compressor

Hydrogen vs Carbon vs Argon vs Flash

ModelRole in the familyIntelligence IndexSpeedPrice (input / output, per million tokens)Best for
Gemini 4 HydrogenFast agentic model (new, 12 Nov 2026)51 (high) / 54 (xhigh)2.5× faster than 3.8 Flash$1.50 / $7.50Everyday agentic work: research, coding, long documents
Gemini 4 CarbonMost capable modelHigher than Hydrogen at defaultSlower, heavierHigher than HydrogenThe hardest reasoning, where quality beats speed
Gemini 4 ArgonBalanced model50Between Carbon and HydrogenBetween Carbon and HydrogenGeneral work where you want a balance
Gemini 3.8 FlashPrevious fast modelLowerBaseline (Hydrogen is 2.5× faster)LowerQuick, simple answers — still available, no shutdown date announced

The decision table is short. If you want the best answer and do not care about speed or cost, Carbon is still the top model. If you want a fast answer to an everyday question, Flash still works. Hydrogen is the new middle that covers most real work: near-Argon quality, Flash-beating speed, and a flat price that does not punish long documents.

How to try it free — Search, Gemini app, AI Studio

  1. In Google Search — no setup at all. Open Google Search and ask a question that needs real work, for example "Compare the battery life and price of the three best wireless earbuds under $100, with sources". Hydrogen is rolling out to all Search users, so the answer you get is increasingly powered by it. You do not need to switch anything on.
  2. In the Gemini app — free, automatic. Open the Gemini app (web or mobile) and ask your question as normal. Free users get Hydrogen through automatic model selection: the app routes suitable tasks to it without you picking a model. Try a task with steps, not a one-line question — "Plan a 3-day trip to Lisbon under $800, with a day-by-day plan and a cost table" is a better test than "What is the capital of France?"
  3. In Google AI Studio — for developers, with control. Open Google AI Studio, pick Gemini 4 Hydrogen from the model list, and choose the thinking level: low for quick checks, high (default) for normal work, xhigh for the hardest tasks. This is where you can also test the 1-million-token context with a long document.
  4. In Google Antigravity — paid Pro/Ultra only. The new Orchestrator switches between Carbon and Hydrogen inside one session, using Hydrogen where it is fast enough and Carbon where the task is hardest. This is the only place Hydrogen needs a paid plan.

Worked example: a real task, step by step

Here is a task that shows what "agentic" actually means — and a way to test Hydrogen yourself in two minutes.

The task: "I have a 30-page PDF report on remote-work trends. Summarise the five most important findings, give me one cited number for each, and make a table comparing 2024 and 2026."

StepWhat you doWhat to look for
1. Shrink the PDFRun the file through a PDF compressor first (see the tool box above) so it uploads fastFinal file near your target size, text still selectable
2. Ask in the Gemini appUpload the PDF and paste the task above, word for wordA five-point summary, each point with a number and a source
3. Check one numberPick one cited figure and find it in the original PDF yourselfThe figure matches the source page — this is the self-check that matters
4. Push it onceReply: "Now turn the table into a one-paragraph summary for a busy manager"It reuses its own table instead of starting over — that is the agentic part

If the summary is right on the number you checked, you can trust the shape of the rest — but never skip step 3 on a number you will quote to someone else. A fast model is still a model: it can be confident and wrong in the same sentence.

Myth vs fact: four things people get wrong

MythFact
"Hydrogen replaces Gemini 3.8 Flash."Google calls Hydrogen the upgrade path from Flash, but has announced no Flash shutdown date. Flash is still available.
"It is only for developers."Developers get it in the API and AI Studio — but it is also in Google Search for everyone and in the free Gemini app.
"Free users cannot use it."Free Gemini app users get it through automatic model selection, and all Search users get it as it rolls out. Only Antigravity needs a paid plan.
"A higher Intelligence Index means it is always better."51 vs Argon's 50 is a narrow lead at the default setting. Carbon is still the most capable model for the hardest reasoning — Hydrogen wins on speed and price.

Check your own result before you trust it

Do this 30-second check on any answer Hydrogen gives you that you plan to act on:

  • ☐ The answer cites its sources — and you opened at least one of them.
  • ☐ One number in the answer matches the source you checked. If it does not, re-ask with "show your working" or switch to xhigh thinking.
  • ☐ For code, you ran it (or pasted it into a test) instead of assuming it works.
  • ☐ For a price or a date, the source is dated this month — a model can quote last year's price with total confidence.
  • ☐ You asked a follow-up question instead of accepting the first answer whole — "What did you leave out?" is a good one.

Mistakes to avoid with a new model

  1. Testing it with trivia. "What is 2 + 2?" tells you nothing. Test it with a real multi-step job — that is what it was built for.
  2. Leaving it on xhigh for everything. xhigh is slower and costs more. Use high by default and xhigh only for hard tasks.
  3. Uploading a 40 MB PDF raw. Shrink it first. Long inputs cost tokens and time, and a compressed PDF works just as well for summarising.
  4. Quoting its numbers without checking one. Always verify at least one cited figure against the original source before you pass it on.
  5. Assuming Search answers are all Hydrogen now. The rollout is in progress — some Search answers still come from other models. Judge the answer, not the label.

Frequently asked questions

What is Gemini 4 Hydrogen?

Google's newest AI model, launched 12 November 2026, built for agentic work — long, multi-step tasks like coding, research and turning dense documents into cited reports. It is fast (2.5× faster than Gemini 3.8 Flash) and priced at $1.50 per million input tokens and $7.50 per million output tokens.

Is Gemini 4 Hydrogen free?

Yes, to try: it is rolling out to all Google Search users and to free Gemini app users through automatic model selection. Developers pay the API price above; Google Antigravity needs a paid Pro or Ultra plan.

How is Hydrogen different from Gemini 4 Carbon?

Carbon is the most capable model in the family — best for the hardest reasoning. Hydrogen is the fast, efficient one: near-Argon quality (51 vs 50 on the Intelligence Index), much quicker, and cheaper. Use Carbon when quality is everything; use Hydrogen for most everyday work.

What is xhigh thinking?

The new top thinking level. Hydrogen has four: low, medium, high (default) and xhigh. xhigh makes the model work harder before answering — it reaches 54 on the Intelligence Index — but it is slower and costs more, so save it for genuinely hard tasks.

Does Hydrogen replace Gemini 3.8 Flash?

No — Google describes Hydrogen as the upgrade path from Flash, but has announced no Flash shutdown date. Flash remains available.

Where can I use Gemini 4 Hydrogen?

Google Search (all users), the Gemini app (free, automatic), the Gemini API and Google AI Studio (all tiers), Gemini for business, and Google Antigravity (paid Pro/Ultra subscribers).

Sources and verification

All launch facts (12 November 2026 launch date; $1.50 / $7.50 per million tokens, flat across the 1-million-token context; Intelligence Index 51 at "high" thinking, 54 at "xhigh", vs 50 for Argon; 2.5× faster than Gemini 3.8 Flash; Terminal-Bench 4.1 score 57% vs 18.2%; four thinking levels low / medium / high / xhigh; rollout to Gemini API, AI Studio, Gemini app, Google Search, Gemini for business and Antigravity Pro/Ultra; Orchestrator results — near-Carbon success with 24% less quota and 19% faster on Terminal-Bench 4.0) are from Google's official blog post "Introducing Gemini 4 Hydrogen for Agentic Workflows" by Koray Kavukcuoglu (SVP, Google DeepMind), 12 November 2026, read in full today. The claim that Hydrogen is available in GitHub Copilot was reported by one outlet on 13 November and has not been independently verified — it is not used in this post. Checked on 16 November 2026. Re-check trigger: Google's next Gemini model announcement or a pricing change on the official blog — the launch post always wins over any guide, including this one.

Your next step: open the Gemini app and give Hydrogen a real multi-step job — the worked example above takes two minutes. If you want to see what Anthropic launched in the same week, our Claude Sheets guide walks through the spreadsheet Claude builds for you, step by step.

About the author
Omkar Tilekar builds MeraToolkit, a free tools website for India — exam form photo and signature resizers, PDF tools, student tools and money calculators that run entirely in your browser. This blog is its guide companion: every figure is checked against the official source before it is published.

Comments