Google launches Gemini 4 Hydrogen for faster coding and agentic workflows

A new model for speed

Google has expanded its frontier lineup with the launch of Gemini 4 Hydrogen. This new model arrives alongside Gemini 4 Carbon and Gemini 4 Argon to provide a diverse range of capabilities for different user needs. While Carbon and Argon are built for deep, long-horizon reasoning on complex problems, Hydrogen is specifically engineered for rapid execution.

The primary goal for Hydrogen is to handle everyday developer workloads with high efficiency. It is designed to perform routine tasks such as writing code, running tests, and verifying changes. By focusing on speed, it serves as a practical tool for software development and other tasks that require quick turnarounds rather than deep philosophical analysis.

According to interestingengineering.com, Hydrogen also excels at broader knowledge work beyond just programming. It can process various inputs including text, images, audio, and video to generate reports or structured outputs. This versatility makes it a flexible addition to the Gemini 4 family for users who need to move quickly through large amounts of data.

Performance and benchmarks

Google highlights significant performance gains for Hydrogen when compared to previous iterations. In evaluations on Terminal-Bench 4.1, the model achieved a score of 57 percent. This represents a substantial leap over Gemini 3.8 Flash, which recorded a score of 18.2 percent on the same benchmark.

Speed is another core metric where Hydrogen stands out for developers. The company reports that the model completes tasks 2.5 times faster than its predecessor when using the default high thinking setting. This speed is intended to help developers maintain momentum during debugging, web development, and multi-step agentic tasks.

The model also shows improved reliability in practical environments. Google notes enhancements in coding accuracy and the model's ability to recover from errors. Specifically, it is better at handling situations where a tool call fails or a search query returns no results, making it more robust for automated workflows.

Adjustable reasoning levels

One of the unique features of Gemini 4 Hydrogen is its ability to balance complexity with cost. It offers four distinct thinking levels: low, medium, high, and a new xhigh setting. This allows users to manually adjust how much reasoning depth the model applies based on the specific requirements of a task.

By providing these levels, Google gives developers more control over latency and expenses. A simple task might only require a low thinking level to save on costs and time, while a more intricate coding problem might necessitate the xhigh setting. This flexibility is key for scaling AI usage across large projects where every token counts.

Hydrogen also supports a memory capacity of up to 1 million tokens for both inputs and outputs combined. This large window allows the model to keep a significant amount of information in its active memory at once. It can analyze large documents, spreadsheets, and charts to produce cohesive summaries or structured data for the user.

The Orchestrator system

To maximize efficiency, Google is introducing a new tool called Orchestrator within its Antigravity development platform. This system allows developers to combine Gemini 4 Carbon and Gemini 4 Hydrogen within a single session. Instead of choosing one model for an entire project, Orchestrator dynamically switches between the two based on the immediate needs of the work.

In this setup, Carbon handles the heavy lifting of initial analysis. It can examine a codebase, resolve ambiguities, and create a comprehensive execution plan. Once the plan is established, Hydrogen takes over to perform the routine coding, package updates, and testing tasks that require high speed.

If Hydrogen encounters a problem it cannot solve, the task can be routed back to Carbon for further reasoning. Google reports that this approach achieves success rates close to Carbon's while using 24 percent less usage quota. On average, this method completes tasks 19 percent faster by avoiding the use of a high-reasoning model for every single step.

Availability and pricing

Google is rolling out Gemini 4 Hydrogen across a wide array of products and platforms. Users can access the model through the Gemini app, Google Search, and the Gemini API. It is also available via Gemini for business and Google AI Studio. It is also available to paid Google AI Pro and Ultra subscribers within the Antigravity platform.

For those using the API, Google has established clear pricing for the 1-million-token context window. The cost is set at $1.50 for every million input tokens and $7.50 for every million output tokens. These prices are designed to help businesses integrate the model into their own software and workflows predictably.

While Hydrogen is broadly available, the Orchestrator tool is initially restricted to Google AI Ultra subscribers. This staged rollout allows Google to monitor how developers interact with the multi-model switching system. As the technology matures, it may become more widely available to the general user base.