Short answer: Gemini 4 Hydrogen is Google's fast, lower cost Gemini 4 model, released on November 12, 2026. It's built for everyday coding and agent work, and unlike Carbon and Argon, it reached almost everyone on day one: the free and paid tiers of the Gemini API and Google AI Studio, the Gemini app, and Google Search (Google). On the paid API tier it costs $1.50 per million input tokens and $7.50 per million output tokens. It also launched with Orchestrator, an option in Google Antigravity where a bigger Gemini 4 model plans and Hydrogen does the work.
If you just saw "Hydrogen" in a model menu and wondered whether to pick it, this guide is for you. It covers what Hydrogen is for, where you can use it and on which plan, what it costs, its four thinking levels, how it compares with Argon and Carbon, and when Orchestrator makes sense.
What is Gemini 4 Hydrogen?
Gemini 4 Hydrogen is the third model in Google's Gemini 4 family. Carbon and Argon are the heavy thinkers, made for long and hard reasoning. Hydrogen is the fast one. Google says it's built "for rapid execution across diverse agentic workflows": writing code, running tests, calling tools and checking results.
Think of it as the model that takes over the slot Gemini Flash used to fill. Google's developer docs call it the recommended upgrade from Gemini 3.8 Flash and earlier Gemini 3 and 2.5 models (Google AI Studio).
A few specs from those docs:
- Model ID:
gemini-4-hydrogen - Context window: one shared window of 1,048,576 tokens for input, thinking and output, the same as Carbon and Argon
- Output: up to 65,536 tokens per request by default, and longer outputs up to the full window with background execution
- Inputs: text, tables, charts, PDFs, images, audio and video
Is Gemini 4 Hydrogen free, and where can you use it?
Yes, in more places than any Gemini 4 model before it. Here's where Google says it's available as of November 12, 2026:
| Where | Who gets access |
|---|---|
| Gemini API and Google AI Studio | Free and paid tiers. The free tier is rate limited |
| Gemini app | Everyone. Free users get Hydrogen through auto model selection, when it's the right model for the prompt |
| Google Search | All users |
| Gemini in Google Sheets | Starting with Google AI Pro and Ultra subscribers |
| Gemini for business | Enterprises, through the API or the model menu |
| Google Antigravity | Paid Google AI Pro and Ultra subscribers, per Google's launch post |
One detail on Antigravity: Google's launch post names paid Pro and Ultra subscribers, while the Antigravity models page also lists Hydrogen for its Free and Google AI Plus plans as of November 13, 2026. If you're on a free plan, check the model menu in your own account.
On the free Gemini app, you don't pick Hydrogen yourself. Auto model selection decides when to use it.
How much does Gemini 4 Hydrogen cost?
On the free tier of the Gemini API, it costs nothing, with rate limits. On the paid tier, Google charges a flat price across the whole 1 million token window:
| Token type | Price per million tokens |
|---|---|
| Input | $1.50 |
| Cached input | $0.075 (95% off, through automatic caching) |
| Output, including thinking tokens | $7.50 |
For comparison, Argon launched at $2 per million input tokens and $10 per million output tokens, an introductory price that Google says ends on November 30, 2026, when Argon moves to $4 and $20. So Hydrogen is cheaper per token. Keep in mind that thinking tokens count as output, so a higher thinking level means a bigger bill for the same question.
What are Hydrogen's four thinking levels?
The thinking level is how much the model reasons before it answers. Hydrogen has four:
| Level | What it's for, per Google |
|---|---|
low | Fast chat and high volume jobs where cost and speed matter most |
medium | Interactive coding, live UI prototyping and quick tool calls |
high (default) | Most tasks, including complex coding and agent work |
xhigh | Hard, multistep problems where high falls short |
xhigh is new with Hydrogen. Google's advice is to start with high and save xhigh for the hard cases, because it uses more tokens, takes longer and can overthink simple prompts. In Antigravity, xhigh shows up as "Extra" in the model menu.
How does Hydrogen compare with Argon and Carbon?
The surprise is that the fast model scores higher than Argon on one widely cited index. Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high setting, ahead of Argon's 50, and 54 at xhigh. That's Google's own reported figure. It doesn't make Hydrogen a replacement for the bigger models on every task.
| Gemini 4 Hydrogen | Gemini 4 Argon | Gemini 4 Carbon | |
|---|---|---|---|
| Role, per Google | Fast execution for everyday coding and agents | Most advanced long tasks, with fewer tokens than Carbon | Google's strongest model, for the hardest problems |
| API price per million tokens | $1.50 in, $7.50 out | $2 in, $10 out until November 30, 2026, then $4 and $20 | Same as Argon |
| Thinking levels | low, medium, high (default), xhigh | Default medium, up to high | Default medium, up to high |
Google also says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting and makes fewer than a fifth as many fabricated claims. Those are vendor numbers too. Run your own tasks before you switch a production workload.
Want the full story on the bigger models? Our guide to Gemini 4 Argon covers Argon and Carbon, including who can use them and how Argon handles security work.
What is Orchestrator in Google Antigravity?
Orchestrator is a new option in Antigravity's model menu that pairs two Gemini 4 models in one conversation. A planner model reads your code, settles the open questions and writes a plan. Then Hydrogen takes over to edit files, write code, run builds and run tests. If Hydrogen gets stuck, say a test fails twice, it hands control back to the planner, which fixes the plan and hands it back (Antigravity docs).
There are two versions:
| Option | Planner | Who does the work | Available on |
|---|---|---|---|
| Orchestrator (Carbon) | Gemini 4 Carbon | Gemini 4 Hydrogen | Google AI Ultra |
| Orchestrator (Argon) | Gemini 4 Argon | Gemini 4 Hydrogen | Google AI Ultra, and paid Google AI Pro plans or trials on eligible devices |
Google says Orchestrator came close to Carbon's success rate on a terminal coding benchmark while using 24% less quota and finishing 19% faster on average. You can't set a thinking level for Orchestrator. It runs with a fixed setup.
Antigravity's docs call Orchestrator the recommended default for everyday feature work, bug fixes and refactoring. If you want a deeper look at how coding agents compare, read our roundup of the best AI coding agents.
Which Gemini 4 should you use?
Here's a simple way to choose, based on Google's own guidance:
| If you want to... | Pick |
|---|---|
| Try Gemini 4 for free, in the app or the API | Hydrogen |
| Do quick coding chores: package upgrades, boilerplate, unit tests, scripts | Hydrogen |
| Build a fast agent or a chat feature on a budget | Hydrogen at low or medium |
| Do everyday feature work in Antigravity with a plan and a review | Orchestrator (Argon or Carbon) |
| Handle long, advanced tasks with fewer tokens than Carbon | Argon |
| Take on the hardest debugging, design or security work | Carbon |
If you're not sure, start with Hydrogen at the default high level. It's the cheapest way in, and you'll quickly see whether a task needs more.
How to try Gemini 4 Hydrogen today
- In the Gemini app. Pick it from the model menu if your plan shows it, or let auto model selection use it.
- In Google AI Studio. Choose
gemini-4-hydrogenand test your prompts on the free tier. - In the Gemini API. Change your model string to
gemini-4-hydrogen. If you're moving from Gemini 3.8 Flash, Google's migration notes say to raise any low output caps, droptemperature,top_pandtop_k, and remove explicit context caching. - In Google Antigravity. Open the model menu and choose Gemini 4 Hydrogen, or Orchestrator if your plan includes it.
Whatever you build, review what the model writes before it ships. Our guide on how to review and verify AI generated code covers the checks that matter.
The skill behind picking a model
New models arrive every few weeks. The names change, but the skill stays the same: knowing which model fits a job, testing it on real work and checking what it produces. If you want to get better at building with AI models and agents, at your own pace, the form below is the place to ask how.
Sources, checked November 13, 2026: Introducing Gemini 4 Hydrogen (Google, November 12, 2026), What's new in Gemini 4 Hydrogen (Google AI Studio), Choose a model (Google Antigravity docs), Gemini 4 Hydrogen and Orchestrator (Antigravity blog, November 12, 2026).