How to Use Gemini 4 Hydrogen: App, Search, AI Studio, API
Gemini 4 Hydrogen launched on 12 November 2026. Where to find it, how to pick it in the Gemini app, which thinking level to use, and how to call it in the API.
Short answer: in Google Search you are already using it, with nothing to switch on. In the Gemini app, free users get it only when automatic model selection decides it fits the prompt; with a Google AI Plus, Pro or Ultra plan you tap the model name at the bottom of the chat box and choose Hydrogen. Developers can try it free in Google AI Studio or call it in the Gemini API as gemini-4-hydrogen. Google announced Gemini 4 Hydrogen on 12 November 2026 and said it was rolling out globally the same day.
Everything below was checked against Google’s own pages on 14 November 2026. Plans, limits and labels change often, so where a step depends on your plan, it says so.
What Hydrogen is for
Hydrogen is the fast, cheaper member of the Gemini 4 family. Carbon and Argon are the heavy models for long, hard reasoning; Hydrogen replaces Gemini 3.8 Flash and is built for getting things done quickly: drafting, research, data analysis, coding fixes and turning a pile of PDFs or spreadsheets into a short cited summary. If you want the family background first, what Gemini 4 Argon is covers the bigger sibling.
A useful rule of thumb: start with Hydrogen. Move up to Argon or Carbon only when a task needs many careful steps and Hydrogen keeps getting the logic wrong. For most everyday work, a faster answer you check yourself beats a slower one you also have to check.
In Google Search
Google says Hydrogen is now available globally in Search for all users, across every language AI Mode supports. There is no toggle and no model name on screen. Ask a question in AI Mode the way you would ask a colleague, with the detail that matters, and Hydrogen answers. The practical change is that longer, more specific questions now get more useful answers, so it is worth writing a full sentence instead of three keywords.
In the Gemini app
This is where your plan decides what you see. According to Google’s Gemini Apps help page on models, as of 14 November 2026:
- No Google AI plan: you cannot pick Hydrogen by hand. With automatic model selection on, Gemini routes some prompts to it. With it off, everything uses Flash-Lite.
- Google AI Plus: Flash-Lite and Hydrogen in chats.
- Google AI Pro: Hydrogen in chats and in task mode, alongside Pro and Argon.
- Google AI Ultra: Hydrogen in chats and task mode, alongside Argon and Carbon.
- Open a chat and tap the model name shown at the bottom of the chat box.
- Choose Hydrogen. Models with a lock icon need a higher plan.
- Optionally pick a thinking level: Low, Medium or High. If you do not choose one, the app uses Low.
- To turn automatic model selection on or off, go to Settings, then Usage limits.
When a paid plan runs out of its usage allowance, the help page says the chat continues on Flash-Lite. If answers suddenly feel thinner late in the day, that is the likely reason.
Which thinking level to use
Thinking levels trade speed for care. In the app you get Low, Medium and High. In the API there is a fourth, xhigh, and the API default is high, not low. A simple way to choose:
- Low: rewording, short emails, summaries of something you pasted, quick factual questions you will check anyway.
- Medium: a plan with a few steps, comparing two documents, a formula that needs explaining.
- High: anything with numbers, code you will run, or an answer that has to hold up in a meeting.
- xhigh (API only): long debugging or analysis jobs where waiting is fine. Google’s model card warns the higher levels can be slow or time out.
A worked example
The prompt matters more than the level. Here is the same request written twice.
Before: Summarise these three supplier quotes.
After: Here are three supplier quotes for 500 office chairs. Put them in a table with unit price, delivery date and warranty. Flag anything that is missing or unclear, and say which quote is cheapest over three years. Do not guess numbers that are not in the quotes.
The second version tells Hydrogen what the output should look like, what to compare and what not to invent. That last line matters: Google reports fewer fabricated claims than Gemini 3.8 Flash, but fewer is not none. Write a prompt that works first try walks through the same pattern for other tasks, and deep research prompts covers longer research jobs.
In Google AI Studio and the Gemini API
Google’s Gemini API what’s new guide for Hydrogen says the model is on both the free and paid tiers of AI Studio and the Gemini API, with the free tier rate-limited. The details a developer needs:
- Model ID: gemini-4-hydrogen. Coming from Gemini 3.8 Flash, change the model string and test.
- Thinking: set thinking_level inside generation_config to low, medium, high or xhigh. Hydrogen rejects minimal with an error.
- Remove temperature, top_p and top_k from your requests, which the guide tells you to drop when migrating.
- Explicit caching is not supported. Implicit caching applies to prompts of at least 4,096 tokens.
- Input, thinking and output share one 1,048,576-token window. Synchronous output stops at 65,536 tokens; longer jobs need background mode.
- Price: $1.50 per million input tokens and $7.50 per million output tokens, in US dollars, with thinking tokens billed as output. Cached input is $0.075 per million.
Google’s developer post adds the use cases it had in mind: bug fixes and small, focused code edits, codebase questions, live UI prototypes, and reports built from PDFs, charts, audio and video. One data point before you paste anything sensitive: on the free tier, Google may use prompts to improve its products. Is Gemini 4 Hydrogen safe explains the paid and free rules, including the different treatment in Europe and the UK.
In Google Antigravity
Google’s coding environment, Antigravity, has Hydrogen for paid Google AI Pro and Ultra subscribers, per the Antigravity team’s launch post. Pick it in the model selector to use it directly, with the same four thinking levels as the API. There is also a new Orchestrator option, where Carbon or Argon plans the work and hands the edits to Hydrogen. Orchestrator with Carbon is Ultra only; with Argon it is on Ultra and paid Pro. AI agent orchestration explains why splitting planning from doing tends to work.
What it still gets wrong
The Gemini 4 Hydrogen model card is plain about it. Hydrogen can still make things up, its knowledge stops in March 2026, and it refuses harmless requests more often than the model it replaces, especially with images. If it declines something reasonable, add context about why you are asking and try again. If it states a figure, ask where the figure came from. How to check an AI answer when you are not the expert is the habit that matters most here.
Where to start on Monday
- Pick one task you already do every week, such as a status update or a summary of meeting notes.
- Run it through Hydrogen at Low, then at High, with the same detailed prompt, and compare.
- Keep the prompt that worked as a template, and note what you had to correct.
- Only move to Argon or Carbon for the tasks where High still falls short.
Coursium is an iOS app that teaches people to use AI at work: writing clear instructions, checking what comes back and applying it to everyday tasks. The model behind the chat box will change again within months; those habits carry over. If you want a feel for the practice first, try the free prompt example, or have a look at Coursium.
Frequently asked questions
Is Gemini 4 Hydrogen free?
Partly. It powers AI answers in Google Search for all users, and it is on the free, rate-limited tier of Google AI Studio and the Gemini API. In the Gemini app, free users only get it through automatic model selection; choosing it by hand needs a Google AI Plus, Pro or Ultra plan.
How do I select Gemini 4 Hydrogen in the Gemini app?
With a Google AI plan, tap the model name at the bottom of the chat box and choose Hydrogen, then optionally a thinking level. Without a plan, turn on automatic model selection under Settings, Usage limits, and Gemini will route suitable prompts to it.
What is the Gemini 4 Hydrogen model ID?
gemini-4-hydrogen. Thinking is set with thinking_level inside generation_config, to low, medium, high (the default) or xhigh.
Which thinking level should I use?
Low for quick rewrites and summaries, Medium for multi-step plans, High for numbers, code and anything that has to be right. In the API, xhigh is available for long jobs where speed does not matter.