Gemini 4 Hydrogen is Google's new fast, everyday model, released on November 12, 2026 as the replacement for Gemini 3.8 Flash. It is free on the Gemini API free tier and in Google AI Studio, costs $1.50 per million input tokens and $7.50 per million output tokens on the paid API, is the new default model in AI Mode in Google Search, and can be picked by name in the Gemini app from the $4.99 Google AI Plus plan up. Free app users only get it when automatic model selection decides a prompt needs it.
The surprise is that the "small" model is not small on scores: on the Artificial Analysis Intelligence Index it scores 51 at its default setting, one point above Gemini 4 Argon. It is still not a replacement for Argon or Carbon on the hardest work, and it costs twice what 3.8 Flash costs until the end of the year.
Facts below are checked against Google's announcement, the Gemini API docs, the model card, Google's plans page and Artificial Analysis on November 13, 2026. We have not run Hydrogen ourselves.
What is Gemini 4 Hydrogen?
Gemini 4 Hydrogen is the third model in Google's Gemini 4 family and the first one meant for everyone. Koray Kavukcuoglu, senior vice president at Google DeepMind and Google's chief AI architect, announced it in a Google blog post on November 12. Google describes Carbon and Argon as its frontier models for deep, long reasoning, and Hydrogen as the model "built for rapid execution" in coding and agent work.
The Gemini 4 Hydrogen model card from Google DeepMind says the model is adapted from Gemini 4 Carbon. It takes text, images, audio and video, and shares one context window of about 1 million tokens across input, thinking and output. A single response can run to 65,536 tokens, and the API can keep generating past that, up to the full window, in the background.
Hydrogen also drops the old "Flash" name. 9to5Google and Ars Technica both describe it as the successor to Gemini 3.8 Flash, and Google's API notes call it the recommended upgrade from 3.8 Flash and older Gemini models.
Google's headline claims against 3.8 Flash: tasks finish 2.5 times faster at the default setting, the Terminal-Bench 4.1 score rises from 18.2% to 57%, and fabricated claims fall by more than five times. These are Google's own numbers. The independent check so far is Artificial Analysis, covered below.
Who can use Gemini 4 Hydrogen?
Unlike Argon and Carbon, which went first to vetted cyber defenders, Hydrogen reached every surface on day one. Access still depends on where you use it and what you pay.
| Where | Free users | Paid users |
|---|---|---|
| Gemini app | Only through automatic model selection; most prompts still go to Flash-Lite | Pick it by name on Google AI Plus ($4.99), Pro ($19.99) and Ultra ($99.99 or $199.99) |
| Google Search | Default model in AI Mode for all users; used for the hardest AI Overviews questions | Same |
| Gemini API and Google AI Studio | Free tier, rate limited | $1.50 input and $7.50 output per million tokens |
| Google Antigravity | No | Google AI Pro and Ultra |
| Gemini in Google Sheets | No | Google AI Pro and Ultra, plus eligible business plans |
| Gemini for business | Not applicable | API and model menu |
The Gemini app rules come from Google's plans page, which now lists "Access to Gemini 4 Hydrogen" as the main model benefit of Google AI Plus, and from the Gemini Apps Help model table as reported by Notebookcheck. If you are on the free plan, leave automatic model selection on. With it off, every reply comes from Flash-Lite and you never reach Hydrogen.
One open question: the plans page no longer lists the Pro model under AI Plus, while the help page still did on November 13. Notebookcheck reports that Google emailed Plus subscribers that they would lose the Pro models on November 13. Check the model menu in your own account. For a side by side of the paid tiers across labs, see ChatGPT Plus vs Claude Pro vs Google AI Pro.
Gemini 4 Hydrogen price
On the API, Google charges one flat rate across the whole context window, with no long prompt surcharge. All figures below are per million tokens, from the Gemini API pricing page on November 13.
| Tier | Input | Output (including thinking) | Cached input |
|---|---|---|---|
| Free tier | Free | Free | Free |
| Standard | $1.50 | $7.50 | $0.075 |
| Batch and Flex | $0.75 | $3.75 | $0.0375 |
| Priority | $2.70 | $13.50 | $0.135 |
Hydrogen supports only implicit caching (repeated prompt prefixes of 4,096 tokens or more get a 95% discount), and Google bills every token processed, including thinking. On the free tier Google may use your prompts to improve its products; on the paid tier it does not.
Against its predecessor, Hydrogen is the pricier model for now. Gemini 3.8 Flash costs $0.75 input and $3.75 output through December 31, 2026, then $1.50 and $7.50 from January 1, 2027, the same as Hydrogen. VentureBeat notes that Google lists no shutdown date for 3.8 Flash yet.
It is cheaper than Argon and Carbon ($2 and $10 until November 30, then $4 and $20), but far dearer per token than other labs' fast models: Claude Haiku 5.5 and GPT-6 Luna start at $0.10 and $0.50, as our Claude Haiku 5.5 vs GPT-6 Luna guide shows. Hydrogen is priced like a midsize model, not a budget one.
Gemini 4 Hydrogen vs Argon, Carbon and 3.8 Flash
| Item | Gemini 4 Hydrogen | Gemini 4 Argon | Gemini 4 Carbon | Gemini 3.8 Flash |
|---|---|---|---|---|
| Role | Fast everyday model | Frontier, lower token use | Strongest frontier model | Previous fast model |
| Public release | Nov 12, 2026 | Oct 13, 2026 | Nov 10, 2026 | September 2026 |
| API price (input / output) | $1.50 / $7.50 | $2 / $10, then $4 / $20 Dec 1 | $2 / $10, then $4 / $20 Dec 1 | $0.75 / $3.75, then $1.50 / $7.50 Jan 1 |
| API free tier | Yes | No | No | Yes |
| Thinking levels | low, medium, high (default), xhigh | up to high (default medium) | up to high (default medium) | default medium |
| Gemini app | Plus and up; free via auto | Pro and Ultra | Ultra | Replaced by Hydrogen |
The Artificial Analysis page for Hydrogen lists 54 on its Intelligence Index at xhigh; Google and VentureBeat give 51 at the default high, against 50 for Argon and 31 for 3.8 Flash. Carbon still leads the Gemini family on that index, as our Gemini 4 Carbon explainer covers. How that index is built is a topic for our sister site's Artificial Analysis Intelligence Index guide.
Two caveats. Hydrogen is wordy at xhigh (304 million output tokens to run the index, far above its price class median), so VentureBeat puts the cost per index task at about $1.87 at high versus $3.16 at xhigh. And Google's own model card shows Claude Sonnet 5.5 ahead on Terminal-Bench 4.1 (63.9% versus 57.0%) and on OSWorld computer use.
Google's pitch is to use them together: a new Orchestrator in Google Antigravity switches between Carbon and Hydrogen in one session, and Google says it came close to Carbon's success rate on Terminal-Bench 4.0 while using 24% less quota and finishing 19% faster.
What changes in Google Search
For most people, Search is where they will meet Hydrogen first. Google's Search VP Robby Stein called it "the biggest model upgrade to Search yet" in the Search announcement. Hydrogen is now the default model in AI Mode for all users worldwide and handles the hardest questions in AI Overviews, as The Verge also reported.
Google says you should notice better handling of long, multipart questions and typos, more visual answers, and a wider "query fan out" that searches subtopics at once and links to more sites. Google has published no counts for the extra links, so treat that as a claim to watch.
If you mostly want answers with sources, our ChatGPT vs Google Search guide compares the two approaches.
What Google claims about Hydrogen's safety
Hydrogen reaches far more people than Argon or Carbon, so its safety claims matter more. Google says that, compared with 3.8 Flash, Hydrogen:
- reduces overreach in agent tasks by up to 81% and cuts misaligned actions and task shortcuts across tool use and coding tests by 12 times;
- is up to two times less likely to bypass permission boundaries when a task is impossible or blocked, and scores 96.7% on Google's internal agent honesty test;
- keeps 99.9% of its reasoning readable, with no deceptive edits to its reasoning found across more than 200,000 audited turns;
- is more than 2.3 times more resistant on the Gray Swan indirect prompt injection benchmark;
- keeps false refusals on HealthBench Pro under 0.4% while screening real biosecurity risks.
These are Google's own evaluations. No outside safety body had published a review of Hydrogen by November 13. For how the big labs write and test their safety rules, see Preparedness Framework vs Responsible Scaling Policy.
Moving from Gemini 3.8 Flash
If you call 3.8 Flash from your own app, the switch is more than a name change. Google's launch notes say the default thinking level rises from medium to high, the old minimal setting returns an error, explicit context caching is gone, and sampling settings such as temperature are ignored. Google advises testing medium and high on your own tasks and keeping the lowest level that meets your quality bar, since xhigh raises cost and can overthink simple prompts. To compare the AI Studio playground with the consumer app, read Google AI Studio vs Gemini.
Pros and cons of Gemini 4 Hydrogen
Pros
- Free on the Gemini API free tier, unlike Argon and Carbon.
- Scores at Argon's level on Artificial Analysis at its default setting, at a lower token price.
- One flat price across the full 1 million token window, with a 95% discount on cached input.
- Four thinking levels, so you can trade speed for depth per task.
Cons
- Twice the per token price of 3.8 Flash until December 31.
- Far more expensive per token than Claude Haiku 5.5, GPT-6 Luna and other budget models.
- Very wordy at xhigh, which raises the real cost per task.
- Free Gemini app users cannot choose it; they depend on automatic selection.
Which Gemini model should you use?
- Free Gemini app user: keep automatic model selection on and Hydrogen will handle the prompts Google judges need it.
- Want to pick Hydrogen yourself: Google AI Plus at $4.99 a month is the cheapest plan that lets you choose it.
- Developer on 3.8 Flash: try Hydrogen on the free tier now, then plan the move before January 1, when Flash rises to the same price anyway.
- Hard engineering, research or security work: use Gemini 4 Neon (from November 18), Gemini 4 Argon or Carbon, or let Antigravity's Orchestrator pair Carbon with Hydrogen.
- Cheapest possible tokens: look at Haiku 5.5, GPT-6 Luna or Google's own Flash-Lite instead.
Choosing between labs instead? See ChatGPT vs Claude vs Gemini vs Grok and our Claude and ChatGPT pages, or start with what is Google DeepMind.
FAQ
Is Gemini 4 Hydrogen free?
Partly. It is free on the rate limited Gemini API free tier and in Google AI Studio, and it is the default model in AI Mode in Google Search for all users. In the Gemini app, free users reach it only through automatic model selection; to pick it yourself you need Google AI Plus ($4.99 a month) or higher.
What is the Gemini 4 Hydrogen price on the API?
$1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens, flat across the 1 million token window. Cached input is $0.075, and Batch and Flex are half price at $0.75 and $3.75.
Is Gemini 4 Hydrogen better than Gemini 4 Argon?
On the Artificial Analysis Intelligence Index, Hydrogen at its default high setting scores 51 against Argon's 50, and 54 at xhigh. Google still positions Argon and Carbon for deep, long reasoning on the hardest problems, and Hydrogen for fast execution. Hydrogen is cheaper per token and is on the free tier; Argon is not.
Gemini 4 Hydrogen vs 3.8 Flash: what changed?
Google says Hydrogen finishes tasks 2.5 times faster at its default setting, scores 57% on Terminal-Bench 4.1 against Flash's 18.2%, and makes more than five times fewer fabricated claims. It also defaults to high thinking, adds an xhigh level and drops explicit caching. It costs twice Flash's price until December 31, 2026, and the same from January 1, 2027.
When was Gemini 4 Hydrogen released?
November 12, 2026, globally across the Gemini API, AI Studio, Gemini for business, the Gemini app and Google Search on the same day. Unlike Argon and Carbon, there was no early access period for cyber defenders first.


