Google introduces Gemini 4 Hydrogen, a new model for software development and agentic workflows. Hydrogen combines advanced reasoning capabilities with lower latency and lower costs. According to Google, the model can perform many development tasks much faster than its predecessor, Gemini 3.8 Flash.
Hydrogen joins the recently introduced Gemini 4 Carbon and Gemini 4 Argon models. While those models focus on complex reasoning and long-running tasks, Hydrogen is designed for fast execution. According to Google, the new model can handle development tasks independently or collaborate with more powerful models.
With the default setting for high reasoning power, Hydrogen scores 51 on the Artificial Analysis Intelligence Index, compared to 31 for its predecessor. This means the new model also performs slightly better than Gemini 4 Argon, which scores 50 points. At the highest reasoning power setting, the score rises to 54.
Compared to Gemini 3.8 Flash, Hydrogen completes tasks 2.5 times faster with the default high reasoning ability setting, according to Google’s own tests. On Terminal-Bench 4.1, the score rises from 18.2 to 57 percent. This benchmark evaluates AI agents performing technical tasks.
Multiple models working together
In addition to Hydrogen, Google is introducing Orchestrator into its Antigravity development environment. This system automatically switches between different AI models within a single workflow.
For example, Gemini 4 Carbon can handle the software architecture and complex decisions, while Hydrogen writes code, runs tests, and verifies changes. According to Google, this division of labor would reduce the need to deploy expensive reasoning models for relatively simple tasks. Google reports that Orchestrator completes tasks on average 19 percent faster, using 24 percent less quota than when Carbon is used alone.
Hydrogen is also designed to improve software development by making targeted changes to existing code, rather than unnecessarily modifying files that are irrelevant to the task. Google reports improvements in debugging, navigating code repositories, and handling failed tool calls.
The model supports multimodal processing, including PDFs, spreadsheets, images, audio, and recorded video. It offers a context window of one million tokens and can generate up to one million output tokens.
Four reasoning levels
Google offers four reasoning settings for Hydrogen: low, medium, high, and extra high. Developers can choose between faster responses and more extensive reasoning, depending on the complexity of their tasks.
API pricing starts at $1.50 per million input tokens and $7.50 per million output tokens. These rates apply to the entire one-million-token context window. According to VentureBeat, this makes Hydrogen currently twice as expensive as Gemini 3.8 Flash. However, starting in January 2027, Flash’s rates will be aligned with Hydrogen’s. Compared to cheaper models from companies such as Anthropic, OpenAI, and Meta, Hydrogen is significantly more expensive per token.
Hydrogen is being rolled out globally via Google AI Studio, the Gemini API, Gemini for Business, the Gemini app, and Google Search. It is also available in Google Antigravity for eligible paying subscribers.