Gemini 4 Hydrogen Is Google’s New Everyday AI Model, and It’s Already in Search

Google dropped the “Flash” name, but not the idea. Gemini 4 Hydrogen, announced on November 12, is the new fast, cheaper-to-run model in Google’s lineup, and it is already doing real work: it’s rolling out in the Gemini app, in Google Search for everyone, and to developers through the Gemini API. If you use Gemini on your Android phone, there’s a good chance some of your answers are coming from Hydrogen this week, whether you picked it or not.

It follows Gemini 3.8 Flash, which launched on September 2, and slots in underneath the two frontier models Google shipped this fall, Gemini 4 Argon and Gemini 4 Carbon. The surprise is how little “underneath” now means.

What Gemini 4 Hydrogen actually is

Hydrogen is Google’s everyday workhorse: the model meant to answer quickly, call tools reliably and run long multi-step tasks without burning a pile of compute. Koray Kavukcuoglu, Google DeepMind’s SVP and Google’s Chief AI Architect, framed it as a model that “brings cutting-edge intelligence to everyone” and is “built for rapid execution across diverse agentic workflows.”

The headline claim is speed. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting. It also says Hydrogen cuts fabricated claims by more than five times compared with 3.8 Flash. That second number matters more to most people than any benchmark. A fast model that makes things up is just a fast way to be wrong.

Hydrogen has four thinking levels: low, medium, high (the default) and a new xhigh tier for harder problems. It takes text, images, audio and video, and works with a context window of roughly one million tokens.

Where you’ll see it on your Android phone

The Gemini app

In the Gemini app’s model picker, Hydrogen sits under “Everyday help.” Free users, Google says, also get Hydrogen through automatic model selection whenever it fits the prompt. In practice, that means the free app should feel quicker and sharper on routine requests without you touching a setting.

The rest of the picker, per 9to5Google, now looks like this: 3.5 Flash-Lite for “Quick answers,” 3.1 Pro for “Detailed research” on AI Pro, 4 Argon for “Code & analysis,” and 4 Carbon for “Most complex work.” Yes, that’s a lot of names. Most people should leave it on auto.

Google Search and AI Mode

This is the bigger deal for U.S. users. Robby Stein, VP of Product for Google Search, called Hydrogen “the biggest model upgrade to Search yet,” and it is available to all Search users globally starting now. Google says Search gets better at long, multi-part and typo-filled questions, fans out to more related subtopics (and therefore more websites), and returns more images, comparison tables and structured layouts. The changes apply across every language AI Mode supports. Google’s post doesn’t say whether AI Overviews specifically are affected, so treat this as an AI Mode story until Google says otherwise.

Everywhere else

Hydrogen is also in Google AI Studio for all tiers, the Gemini API, Android Studio, Gemini for business, and Gemini in Google Sheets, where Google says it generates outputs over 30% faster than previous models. Google’s Antigravity coding tool gets it for paid Google AI Pro and Ultra subscribers.

How Hydrogen stacks up

On the independent Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high setting. That’s one point ahead of Gemini 4 Argon, Google’s own frontier model from October, and a 20-point jump over Gemini 3.8 Flash. At xhigh it reaches 54.

Google’s own coding numbers are even more lopsided. On Terminal-Bench, Hydrogen completes 57% of tasks against 18.2% for 3.8 Flash. It isn’t the class leader everywhere, though: VentureBeat notes Anthropic’s Claude Sonnet 5.5, a pricier mid-tier model, scored 63.9% on the same test. Google’s materials also cite the benchmark as both version 4.0 and 4.1 in different places, so take the exact comparison with a grain of salt.

Business testers report similar gains. Box CEO Aaron Levie said that on a typical task in Box’s internal evaluation, Hydrogen took less than half the time and about 12% fewer tokens, while accuracy rose from 74% with Flash to 79%.

The catch: it’s not cheap anymore

“Flash” used to mean bargain. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, with no introductory discount. Gemini 3.8 Flash is currently $0.75 and $3.75 on a promotional rate that runs through December 31, and it rises to the same $1.50 and $7.50 on January 1, 2027. Rival small models such as Claude Haiku 5.5 and GPT-6 Luna start at $0.10 per million input tokens.

Google’s argument is that Hydrogen spends fewer tokens to finish a job, so the bill per task can still come out lower. Artificial Analysis estimates Hydrogen’s cost per benchmark task at $1.87 at high, versus $7.77 for 3.8 Flash. Batch and Flex processing are half price for developers who can wait.

None of that touches you if you use the free Gemini app or Search. It matters to the Android developers building on the API, and it signals where Google thinks the value is now: not the cheapest possible model, but a mid-priced one that’s smart enough to replace a frontier model for most work.

What it means for you

For most Android owners, the takeaway is simple. You don’t need to do anything. Update the Gemini app from the Play Store if it’s been a while, leave the model picker on auto, and try a long, messy question in AI Mode, the kind that used to need three separate searches. That’s where Hydrogen is supposed to shine.

The more interesting shift is strategic. A model priced as a mid-tier option now edges out Google’s own October flagship on a widely watched index. If you’ve been paying for Google AI Pro mainly to get “the smart model,” Hydrogen narrows that gap for everyday tasks. Pro still buys Carbon for the hardest work, plus Antigravity access, but the free tier just got a lot more capable.

Sources: Google, Google Search blog, VentureBeat, 9to5Google. Photo by Magnet.me on Unsplash.