Sunday, November 15, 2026 Independent reporting on artificial intelligence Newsletter
AI News Beats
Models & Research

Google brings Gemini 4 Hydrogen to developers, Search and its app

The November 12 launch adds a lower-cost Gemini 4 option for application developers while Google rolls the model into its consumer search and assistant products.

An open laptop displaying code beside a coffee mug and plant
Photo: Daniil Komov / Unsplash

Google introduced Gemini 4 Hydrogen on November 12, expanding the latest generation of its AI models beyond the higher-end Carbon and Argon releases. The company is offering Hydrogen to developers through the Gemini API and Google AI Studio and is bringing it to the Gemini app and Google Search.

Hydrogen is aimed at the work developers repeat throughout a project, such as editing a small section of code, reading a test result or navigating a collection of files. Google says it can also be used alongside its more powerful models. The distinction matters to teams that pay not only for the model’s answer but also for every intermediate step it takes to arrive there.

What the new model costs

Google lists paid API rates of $1.50 per million input tokens and $7.50 per million output tokens. Tokens are the units of text and other information a model processes. The company says the same rates apply across the model’s one million-token context window, the amount of information it can consider in a single request. It also lists a free tier for developers.

Those published prices are not an estimate of the cost to complete a project. A workflow that sends a large amount of background information several times, or retries a task after a poor answer, can cost more than a shorter workflow with a higher token rate. Businesses evaluating Hydrogen will need to compare the total cost of a successful run, rather than the list price alone.

Choosing how much reasoning to use

The model has four levels of reasoning effort, with high as its default. A developer can select a lower setting for straightforward work or reserve the highest setting for a harder task. The settings provide another option besides changing models, although the right choice depends on the application’s tolerance for errors and delays.

Google says Hydrogen finished tasks 2.5 times faster than Gemini 3.8 Flash at the new model’s default high setting in its testing. The company also reported a higher score on its cited terminal benchmark. These are Google’s reported comparisons, not a guarantee of a similar improvement for a particular software repository, language or customer workload.

Different ways to access Hydrogen

Developers can work with the model in AI Studio or through the API, while Google lists it for business customers and for paid AI Pro and Ultra users in its Antigravity development environment. The consumer app and Search are separate distribution channels. Their interface, access conditions and built-in tools are not necessarily the same as the API.

For teams already using a Gemini Flash model, the next decision is whether Hydrogen can handle the same tasks without increasing review work. A comparison on existing test cases can track completion time, the number of retries and the proportion of results a person accepts. The launch gives teams a new model to evaluate against those measures.

Stay on top of AI

Get the stories that matter in artificial intelligence, straight to your inbox.

Free. One email with the stories that matter. Unsubscribe any time. See our privacy policy.