Google launches Gemini 4 Hydrogen, its fast model, across the Gemini app, Search and the API

Google released Gemini 4 Hydrogen on Thursday, the speed- and cost-focused member of its Gemini 4 model family. It's rolling out in the Gemini app, to developers through Google AI Studio and the Gemini API, and in Google Search, where Google says it is now available globally for all users. The launch comes a week after Gemini 4 Carbon became available, according to 9to5Google.

A new name for the fast tier

Hydrogen takes over from the "Flash" branding Google used for its earlier quick models; its predecessor was Gemini 3.8 Flash, launched in September. Google pitches it for everyday coding and agent-style tasks where speed and price matter.

The numbers Google is leading with, as reported by 9to5Google and repeated on Google's Antigravity blog:

9to5Google also reports a 58.8% score on HLE-Diamond and first-place or tied-first results on two Vals benchmarks for tax and legal research. Compared with 3.8 Flash, Google says Hydrogen finishes tasks 2.5 times faster at its default setting and makes up facts more than five times less often, and that partners measured runs 2 to 2.8 times faster using 12% to 23% fewer tokens. These are Google's and partners' figures, not independent tests.

The model accepts up to 1 million tokens of input and can produce up to 1 million tokens of output. API pricing is $1.50 per million input tokens and $7.50 per million output tokens, 9to5Google reports.

Where you'll see it

• Gemini app: It powers the "Everyday help" option in the model picker, and free users get it when automatic model selection decides it fits a prompt.
• Google Search: In a post by Search product VP Robby Stein, Google calls this "the biggest model upgrade to Search yet." It says Hydrogen handles multi-part questions and typos better, runs more parallel sub-searches to surface a wider range of links, and produces more images, tables and comparison layouts. Google gives no benchmarks for the Search claims.
• Developers: AI Studio, the Gemini API and Gemini for business.
• Antigravity: Google's coding tool gets Hydrogen for paid AI Pro and Ultra subscribers, plus a new Orchestrator option for Ultra subscribers that pairs it with Carbon. Carbon plans the work in a read-only phase, Hydrogen makes the edits and runs tests, and the task goes back to Carbon if Hydrogen gets stuck. Google says that on Terminal-Bench 4.1, this setup gets close to Carbon's success rate while using about 24% less quota and running about 19% faster.

Why it matters

For most people, the fast tier is the Gemini model they actually use. It answers by default in the app for free users when Google's auto-selection picks it, and it now sits behind Search for everyone, which means Hydrogen's quality will shape what everyday Google users see far more than Google's flagship models will.

The headline claim is that a cheaper, faster model edges out a larger sibling on an independent index, if only by a point. The Orchestrator approach is the other thing to watch: splitting a job between an expensive planner and a cheaper worker is a straightforward way to cut AI costs, and Google is now packaging it as a menu option. As with any vendor benchmark, real-world results on your own tasks are the test that counts.