Gemini 4 Hydrogen is now generally available in the Gemini API, with a free tier and paid pricing of $1.50 per million input tokens and $7.50 per million output tokens. Google released it on November 12, 2026. It adds a fourth thinking level, xhigh, and it is now the default model for the Antigravity agent in Gemini Managed Agents.
Hydrogen is the third Gemini 4 model. We covered the first in Gemini 4 Argon’s general availability and laid out the older lineup in which Gemini model to use. This post explains where Hydrogen fits, what it costs for a small site chatbot, how to move off Gemini 3.8 Flash and what changed in Google Search. Facts are as of November 16, 2026.
The short version: Gemini 4 Hydrogen is the Gemini 4 model most small teams will actually run. It is the only Gemini 4 model with a free tier. Its paid price sits below Argon and Carbon. Google also put it into Search for all users the same day. If you run Gemini 3.8 Flash, plan the move now, but check the math first. Flash is cheaper than Hydrogen until December 31.
What Google released with Gemini 4 Hydrogen
The Gemini API changelog entry for November 12 lists gemini-4-hydrogen as generally available. It names the free tier, the $1.50 and $7.50 paid prices, the new xhigh level and the Antigravity default. It also points to a migration guide from Gemini 3.8 Flash.
Google’s launch post by Koray Kavukcuoglu says Hydrogen is "built for rapid execution across diverse agentic workflows." Google says it is rolling out across the Gemini API, AI Studio, Gemini for business, the Gemini app and Google Search. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity.
The rest of the specs come from Google’s What’s new in Gemini 4 Hydrogen guide:
- Context: one 1,048,576-token window shared by input, thinking and output.
- Output: up to 65,536 tokens per request, and up to 1M with background execution.
- Thinking levels: low, medium, high (the default) and xhigh.
- APIs: both GenerateContent and the Interactions API.
Google also reports results against Gemini 3.8 Flash. It says Hydrogen finishes tasks 2.5 times faster at its default setting and scores 57% on Terminal-Bench 4.1, against 18.2% for 3.8 Flash. It also says Hydrogen cuts fabricated claims by more than 5x. These are Google’s own numbers.
Gemini 4 Hydrogen vs Argon and Carbon
Google ranks the three Gemini 4 models by how hard they think. Its guide calls Carbon "Our strongest frontier model." It says Argon handles most advanced long tasks with fewer tokens than Carbon. Hydrogen is the fast one, recommended for everyday coding, tool-using agents and document work.
The prices below are from the Gemini API pricing page, read November 16, 2026.
| Model | Paid price per 1M tokens (input / output) | Free tier | Google’s suggested use |
|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 / $7.50 | Yes | Everyday coding, agents, document work |
| Gemini 4 Argon | $2 / $10 through Nov 30; $4 / $20 from Dec 1 | No | Most advanced long tasks |
| Gemini 4 Carbon | $2 / $10 through Nov 30; $4 / $20 from Dec 1 | No | Hardest debugging, science and security work |
Carbon is now in the API too. The changelog lists Carbon as generally available on November 10, 2026.
On Google’s own benchmark claim, Hydrogen edges Argon. The launch post says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at high, against 50 for Argon, and 54 at xhigh. Treat that as a vendor claim until your own tests agree.
Google also suggests pairing models. Carbon plans or untangles a hard problem. Hydrogen then does the edits, test runs and parallel subagent work. Switching between them needs no request schema changes, per the guide.
The new xhigh thinking level
Hydrogen has four thinking levels. Argon and Carbon default to medium and stop at high. Hydrogen defaults to high and adds xhigh.
Google’s guide gives plain advice. Start with high for most complex tasks. Save xhigh for hard, multi-step problems where high falls short. The guide warns that xhigh raises latency and token use and "can cause overthinking on simpler prompts."
Thinking tokens bill as output tokens at $7.50 per million. So the level you pick is a cost setting. For a site chatbot that answers simple questions, test low or medium first. Run your evaluations at each level and keep the lowest one that meets your bar.
What Gemini 4 Hydrogen costs for a small site chatbot
Here is a worked example. Say a nonprofit’s help bot sends 4,000 input tokens per answer: instructions, a few retrieved pages and the question. It gets back 800 output tokens, thinking included. These are our assumptions. Your numbers will differ.
| Model and price period | Cost per answer | 10,000 answers a month |
|---|---|---|
| Gemini 4 Hydrogen | $0.012 | $120 |
| Gemini 3.8 Flash, through Dec 31, 2026 | $0.006 | $60 |
| Gemini 3.8 Flash, from Jan 1, 2027 | $0.012 | $120 |
| Gemini 4 Argon, from Dec 1, 2026 | $0.032 | $320 |
| Gemini 3.5 Flash-Lite | $0.0032 | $32 |
The pricing page shows Gemini 3.8 Flash at $0.75 input and $3.75 output through December 31, 2026. From January 1, 2027, it rises to $1.50 and $7.50, the same as Hydrogen. So the price gap closes on New Year’s Day.
Two things can lower the Gemini 4 Hydrogen figure. First, implicit caching. The guide says it runs on prompts of at least 4,096 tokens and cuts cached input by 95%, to $0.075 per million. A long, fixed set of bot instructions qualifies. Our prompt caching explainer covers how to order a prompt so the cache hits. Second, fewer thinking tokens at a lower level.
One more cost note. The free tier is free of charge, but the Gemini API terms say Google uses content sent to unpaid services to improve its products, and human reviewers may read it. Keep member data and anything confidential on the paid tier.
How to migrate from Gemini 3.8 Flash
Google’s guide has a migration checklist. These are the steps most site builds will hit:
- Change the model ID from
gemini-3.8-flashtogemini-4-hydrogen. - Raise low output caps. Hydrogen shares one 1M window across input, thinking and output. Google says to remove hardcoded caps such as 8,192 or raise them to 65,536.
- Set the thinking level. Hydrogen defaults to high, not medium. It rejects
minimalwith an error. - Drop explicit caching. Hydrogen does not support
cachedContentsorcaches.create. Removecached_contentfields and rely on implicit caching. - Remove sampling parameters. Delete
temperature,top_pandtop_k. Out-of-range values return a 400 error. - Declare tools natively. Do not ask for raw JSON tool calls in the system prompt with no declared tools. That can return
MALFORMED_FUNCTION_CALL. - Keep thought signatures. In multi-turn calls, return prior model turns unchanged, or use
previous_interaction_idin the Interactions API.
Our post on Gemini 3.8 Flash’s cost per task showed why token counts matter more than list prices. Measure tokens per answer on both models before you switch a busy bot.
The Antigravity agent now defaults to Gemini 4 Hydrogen
If you call the Antigravity agent through Gemini Managed Agents, your model may have changed without a code change. Google’s Antigravity agent guide says that for antigravity-preview-09-2026, "If you omit agent_config, the agent defaults to gemini-4-hydrogen."
You can still pick the model. Set agent_config to gemini-4-carbon for harder engineering work, per the same guide. Check your logs for token counts after November 12. We covered the agent setup in the Antigravity SDK and Managed Agents.
What Gemini 4 Hydrogen changes in Google Search
Google’s Search post says this is "the biggest model upgrade to Search yet." According to Robby Stein’s November 12 post, Hydrogen is now available globally in Search for all users.
Google names four changes:
- Intent: Search handles multi-part questions, typos and specific requests better.
- Query fan-out: Google says Search runs simultaneous queries across subtopics and links to "a broader range of helpful links than before."
- Visual answers: more images, comparison tables and custom tools.
- Languages: better answers across all supported AI Mode languages.
The fan-out change is the part to watch. Google says people may find sites they "previously might not have discovered." That could mean more citations for a focused page on a narrow question. It could also spread clicks thinner. We tracked Google’s earlier link changes in more links in AI Mode and AI Overviews.
Google’s two posts do not say which AI Mode or AI Overviews answers run on Gemini 4 Hydrogen, or whether other models still handle some queries. Watch Search Console for changes in AI traffic and note the date, rather than assume a cause.
Similar moves at OpenAI and Anthropic
All three labs now sell a fast, lower-cost model as the everyday default. OpenAI’s GPT-6.1 Terra model page lists $1 input and $5 output per million tokens. We covered it in GPT-6.1 Terra. Anthropic’s pricing page lists Claude Sonnet 5.5 at $2 and $10, and Claude Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens. See our post on Claude Haiku 5.5. Google also put its new model into Search on launch day.
What to do this week
- List every place you call Gemini 3.8 Flash. Plugins, chatbots and scheduled jobs count.
- Run Gemini 4 Hydrogen at low, medium and high on twenty real questions. Record tokens per answer.
- Decide on timing. Flash is half Hydrogen’s price until December 31. If quality is equal, you can switch in late December.
- Check free-tier keys. Move any key that sends member or donor data to a paid project.
- Pin Antigravity’s model in
agent_configif you need predictable costs. - Mark November 12 in your analytics so later AI Mode traffic shifts have a date to compare against.
Frequently Asked Questions
What is Gemini 4 Hydrogen?
It is Google’s fast, lower-cost Gemini 4 model, released as generally available on November 12, 2026. Google built it for everyday coding, agents and document work.
How much does Gemini 4 Hydrogen cost?
The paid tier is $1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens. Cached input is $0.075 per million. There is also a rate-limited free tier.
Is the free tier safe for member data?
Google’s Gemini API terms say content sent to unpaid services is used to improve Google products and may be read by human reviewers. Use the paid tier for anything sensitive.
What is the xhigh thinking level?
It is a new fourth level above high. Google says to use it only for hard, multi-step problems, because it raises latency and token use.
Is Hydrogen better than Gemini 4 Argon?
Google reports a score of 51 for Hydrogen at high on the Artificial Analysis Intelligence Index, against 50 for Argon. That is a vendor claim. Google still places Argon and Carbon above Hydrogen for long, deep reasoning.
Should I switch from Gemini 3.8 Flash now?
Test now and decide on timing. Flash costs half as much as Hydrogen through December 31, 2026, and rises to the same price on January 1, 2027.
What breaks when I migrate from 3.8 Flash?
Explicit caching, the minimal thinking level and sampling parameters such as temperature. Low output caps may also cut answers short. Google’s migration checklist lists each change.
Did the Antigravity agent change?
Yes. The antigravity-preview-09-2026 agent now uses gemini-4-hydrogen when you do not set a model in agent_config.
What does Hydrogen change in Google Search?
Google says Search now understands complex questions better, shows more visual answers and links to a wider range of sites through improved query fan-out. It is available to all users globally.