Google introduced Gemini 4 Hydrogen. The nimble model is heading to Search and the Gemini app

The smartest artificial intelligence models can solve complex problems, but you often have to wait for their answers, and their operation is expensive. That’s why Google is introducing Gemini 4 Hydrogen, a lightweight member of the Gemini 4 family, designed to be primarily fast and reliable for everyday tasks. Google is offering it to everyone, from users of the Gemini app and Google Search to developers and businesses.

Hydrogen is a nimble worker, Carbon and Argon do the thinking

The Gemini 4 family now consists of three members. According to Google, the Gemini 4 Argon model, which we wrote about at the beginning of October, and Carbon are designed for deep and extensive reasoning on the most difficult problems. Hydrogen has a different role. As Google states on its blog, it is built for quickly handling everyday tasks: writing texts, research, working with documents, or programming.

The models can also be combined: the powerful one devises the procedure, and the workhorse Hydrogen performs the actual work. In the Google Antigravity developer tool, a new Orchestrator feature automatically switches between Carbon and Hydrogen. In the Terminal-Bench 4.0 programming test, according to the company, it achieved almost the same success rate as Carbon alone, while being 19% faster and consuming 24% less of the user’s available limit.

It is expected to be 2.5x faster than Gemini 3.8 Flash

Hydrogen builds on Gemini 3.8 Flash, and Google promises a big leap. It is expected to complete tasks 2.5 times faster, and in the Terminal-Bench 4.1 test, which measures how well the model handles command-line work on a computer, it improved from 18.2% to 57%. It is also important that, according to Google, it hallucinates more than five times less.

In the comprehensive Artificial Analysis Intelligence Index, compiled from a series of tests, Hydrogen scores 51 points by default, one point more than Argon. At the highest level of reasoning, it reaches 54 points. However, for long and complex reasoning, Google continues to recommend Argon and Carbon.

Available today in the Gemini app and Search

Regular users will primarily encounter Hydrogen in the Gemini app, where they will find it in the model selection. Google promises that it will adhere more precisely to the specified tone, format, and rules, so you will get usable email drafts or study materials right away. It can also retrieve information from uploaded files, Gmail and Google Drive, as well as from connected services like Dropbox or Airtable.

An even more visible change awaits Google Search. Robby Stein, Vice President of Search, on the company blog, speaks of the biggest model improvement in Search’s history. Hydrogen is designed to better understand multi-part queries and even typos, and to respond more clearly, with images, comparison tables, and tools created specifically for your query. At the same time, it is intended to search more related topics for a single question at once, thus offering a wider selection of links. The change applies to all users worldwide and to all languages in AI mode.

Developers can use Hydrogen via the Gemini API and AI Studio and choose from four levels of reasoning. The longer the model thinks about a task, the better the result, but the slower and more expensive the response. Additionally, Google AI Pro and Ultra subscribers receive it in the Antigravity tool and Google Sheets.

Will you let Gemini 4 Hydrogen write your emails, or do you not trust it yet?

Source: Google