Google just launched a third Gemini 4 model. And it’s not the biggest one.
Gemini 4 Hydrogen is built to be fast and cheap. Google calls it its “daily driver for agentic workflows.” The pitch is simple: do more work, spend less, and show up in Search too.
Some summaries inflate the story. Let’s sort what’s confirmed from what’s just Google’s own claim.
Quick answer
What is Gemini 4 Hydrogen? A fast, low-cost Gemini model for agent work, coding and everyday tasks. It replaces Gemini 3.8 Flash.
When did it launch? November 12, 2026.
What does it cost? $1.50 per million input tokens and $7.50 per million output tokens, with a 1 million token context window.
Is it Google’s smartest model? No. Gemini 4 Carbon and Argon are still Google’s deep-reasoning models.
What is it actually for?
Think of Hydrogen as the workhorse. Google positions it as a fast execution layer, not a replacement for its frontier reasoning models.
It has four thinking levels: low, medium, high (the default) and a new xhigh. More thinking means better answers, but more tokens and more cost.
Context is 1 million tokens in, and up to 1 million out. The price stays flat across the whole window.
How does it score?
On the Artificial Analysis Intelligence Index:
- Hydrogen (high): 51
- Hydrogen (xhigh): 54
- Gemini 4 Argon: 50
- Claude Sonnet 5.5 and GPT-6.1 Sol (max): 55 each
So it edges past Argon by one point at its default setting. But it trails the top rivals at their maximum settings. “Beats Argon” is true. “Beats everything” isn’t.
The Decoder points out Argon looks less special now. Google had recommended Argon for daily coding just two days before Hydrogen arrived.
Where is it weak?
Artificial Analysis tests show some gaps:
- Terminal-Bench 4.0: 54%, against 64% for Claude Sonnet 5.5 and 56% for GPT-6.1 Sol at max
- Terminal-Bench-Science: 50%, against 53% and 58%
It also falls short of Carbon on most agentic coding and long-context tests where both are listed. Hallucination rate on one test came in at 8%.
Is it really “13 minutes per task”?
I couldn’t confirm that number. Google’s claim is that Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at the default setting. One summary notes the launch post gives no standalone latency or minutes-per-task figure.
So treat “13 minutes” as unverified.
What does it really cost per task?
A common mix-up. $1.50 and $7.50 are per-million-token rates. They’re not the cost of a task.
Artificial Analysis measured cost per task on its index:
- xhigh: $3.16
- high: $1.87
- medium: $0.80
- low: $0.29
At xhigh, that’s about a third of Claude Sonnet 5.5 at max and about 2.5 times GPT-6.1 Sol at max. At high, medium and low, no other model is both cheaper and smarter, according to The Decoder’s reading of the data.
What is Orchestrator in Antigravity?
Orchestrator is a feature in Google Antigravity, Google’s agent coding platform. It switches between models during one session.
The suggested setup: Carbon coordinates the work. Hydrogen powers the worker and verifier agents that build and test changes in isolated workspaces.
On Terminal-Bench 4.0, Google says this setup came close to Carbon’s success rate while using 24% less quota and finishing 19% faster. That’s Google’s own number. Also note Hydrogen’s headline score uses version 4.1 of that benchmark, so the two figures aren’t directly comparable.
Access is unclear. One summary says Pro and Ultra subscribers. Google’s own tweet reportedly says Orchestrator is rolling out to Ultra.
What changes in Google Search?
Google says Hydrogen is rolling out in Search worldwide, for all supported AI Mode languages. It says the upgrade brings:
- Better handling of typos, multi-part questions and specific requests
- More visual answers
- Broader links, through improved query fan-out
Here’s what I couldn’t find. No source says Hydrogen powers every Google search, or gives a figure like “billions of queries.” The reporting describes AI Mode and the Search experience. Full regional details aren’t published either.
Who can use it?
- Developers: Gemini API in Google AI Studio and Android Studio
- Businesses: Gemini for business, plus Gemini in Google Sheets on eligible plans
- Coders: Google Antigravity, for paid AI Pro and Ultra subscribers
- Consumers: the Gemini app and Google Search. Free users reportedly get it through automatic model selection when it suits the prompt. One report’s headline says Pro and Ultra only, so check your own app.
In the Gemini app, Google says Hydrogen follows tone, format and instructions better over long chats, and can pull context from Gmail, Drive, Airtable and Dropbox.
What about safety claims?
Google reports these versus Gemini 3.8 Flash, from its own testing:
- Up to 81% less agentic overreach
- Up to 79% less under-service
- A 96.7% internal honesty rate
- Zero deceptive reasoning edits across more than 200,000 audited turns
These are self-reported and not independently verified.
Should you use it?
- Yes, test it for high-volume agent steps, tool calls, verification and everyday coding. That’s what it’s built for.
- Stick with Carbon or a top rival for hard, long-horizon engineering.
- Budget by cost per task, not token price. Xhigh uses nearly twice the output tokens of high.
- Run your own tests. Most numbers come from Google.
What we still don’t know
- Real latency, in seconds or minutes per task
- Exact Search regions and languages
- Who gets Orchestrator, and how its routing decisions work
- Independent checks of the safety numbers
FAQ
What is Gemini 4 Hydrogen?
It’s Google’s fast, low-cost Gemini model for agent workflows, coding and everyday tasks. It replaces Gemini 3.8 Flash.
How much does Gemini 4 Hydrogen cost?
$1.50 per million input tokens and $7.50 per million output tokens.
Is Hydrogen better than Gemini 4 Argon?
Slightly, on the Artificial Analysis index: 51 versus 50 at default settings. Carbon and Argon remain Google’s deep-reasoning models.
Is Hydrogen in Google Search?
Google says it’s rolling out in Search worldwide for supported AI Mode languages. It isn’t confirmed that it handles every search.
What is Orchestrator?
A feature in Google Antigravity that switches between Carbon and Hydrogen during a coding session.
Is Hydrogen free?
Free Gemini app users reportedly get it through automatic model selection. API use is paid.
What is its context window?
1 million tokens in, up to 1 million out.
Bottom line
Hydrogen isn’t Google’s smartest model. It’s its most practical one. Fast, cheap and good enough to run the repetitive parts of agent work, while Carbon handles the hard thinking.
The “13 minutes per task” and “billions of searches” lines are not backed up. The real story is cost and speed, and that’s still a big deal.