Google launches Gemini 4 Hydrogen for fast tasks
Google launches Gemini 4 Hydrogen, a model designed for programming, AI agents and everyday tasks with lower latency. It is available globally through Gemini, Search, the API and enterprise products, with a 1 million token context window.

Google has launched Gemini 4 Hydrogen, a model designed to handle programming, research and everyday tasks faster than its deep-reasoning models. It is now available globally through the Gemini API, AI Studio, the Gemini app, Google Search and several enterprise products.
Hydrogen completes the Gemini 4 family alongside Gemini 4 Carbon and Gemini 4 Argon. Carbon and Argon are built for complex problems that require extended analysis, while Hydrogen is designed to execute instructions quickly: writing code, testing changes, consulting files, summarizing documents or coordinating AI agents.
A fast model, but not a limited one
On the Artificial Analysis Intelligence Index, Hydrogen scores 51 compared with Argon's 50 with its high reasoning level enabled by default. At the most demanding level, called xhigh, it reaches 54, according to Google.
The model offers four reasoning levels: low, medium, high and xhigh. The idea is to let you choose between shorter waits and lower costs for simple tasks, or more analysis when a problem requires it.
It also supports a 1 million token context window. That means it can work with large amounts of text, code or information at once. The price is $1.50 per million input tokens and $7.50 per million output tokens, regardless of the window size.
Orchestrator combines multiple models
Google has also introduced Orchestrator within Google Antigravity. This tool switches between models automatically based on the stage of a task instead of using a single model from start to finish.
For example, Carbon can decide a program's architecture while Hydrogen writes the code, runs tests and verifies the changes. In Terminal-Bench 4.0, Google says this combination delivered results close to Carbon's, used 24% less quota on average and completed tasks 19% faster.
In projects with multiple agents, Hydrogen can work as an execution or verification agent. This allows several processes to write and test changes in separate workspaces at the same time.
The jump from Gemini 3.8 Flash
For teams coming from Gemini 3.8 Flash, Google says Hydrogen completes tasks 2.5 times faster with its high default configuration. It also says the model's Terminal-Bench 4.1 score rises from 18.2% to 57% and that it reduces fabricated claims by more than five times.
In development and enterprise work tests, the model is designed to:
- Detect and fix errors without modifying files that are not necessary.
- Read and search multiple files in parallel across large repositories.
- Run research, data analysis and code navigation tasks.
- Turn PDF documents, spreadsheets, charts, audio and video into structured reports.
- Create functional web prototypes with fewer instructions.
In initial tests cited by Google, some partners found that Hydrogen ran certain workflows 2 to 2.8 times faster than 3.8 Flash and reduced token usage by between 12% and 23%.
What changes for Gemini users
In the Gemini app, Hydrogen improves tasks such as drafting emails, preparing study guides, rewriting text or comparing products. Google says it follows tone and formatting instructions more closely and needs fewer exchanges to produce a usable result.
It can also combine information from uploaded files, Gmail, Google Drive and connected services such as Airtable or Dropbox. In practice, you could ask it to gather data from several sources and turn it into a table or document instead of copying the information manually between apps.
In Google Search, Hydrogen powers answers that aim to better interpret long questions, typos and multi-part queries. Google is also increasing the number of relevant links shown by Search so you can find pages that might previously have been left out of the main results.
Safety and availability
Google says Hydrogen reduces agent over-action by up to 81% and cases where agents do less than necessary by up to 79%, according to its internal evaluations. It also says the model is 12 times less likely to take misaligned actions or shortcuts in tool-use and programming tests.
These figures come from Google's tests, not a guarantee for every conversation. The model can still make mistakes, especially when it has to interpret ambiguous instructions or act on external services.
Gemini 4 Hydrogen is beginning its global rollout across several service tiers:
- Developers: Gemini API, AI Studio, Android Studio and Google Antigravity for Google AI Pro or Ultra subscribers.
- Businesses: Gemini for businesses, Google Antigravity and Gemini in Google Sheets for eligible subscriptions.
- General users: Gemini app, Google Search and Gemini in Google Sheets for Google AI Pro and Ultra subscribers.
The important part of this launch is not just that Hydrogen reasons faster. Google is betting on a way of working in which each model handles the part it is best at: one plans, another executes and another checks. The next thing to watch is whether this coordination actually reduces costs and errors when these tools are used in everyday projects, not just benchmarks.

