Google has launched Gemini 4 Hydrogen, an advanced AI model that performs tasks 2.5 times quicker than its earlier version, Gemini 3.8 Flash, as the company aims to enhance the speed and effectiveness of routine AI activities, programming, and agentic processes.

The model will be accessible through Google AI Studio, the Gemini API, Gemini for business, the Gemini app, and Google Search. According to the company, Google Antigravity is also accessible to paid subscribers of Google AI Pro and Ultra.

In a post on X, Sundar Pichai, the CEO of Google, mentioned that the model was designed for rapid performance and effectiveness in routine tasks, coding, and autonomous workflows.

"In relation to 3.8 Flash, it completes jobs 2.5 times quicker." "It also offers 4 levels of reasoning, ensuring you only wait and pay for the depth a task needs," he mentioned.

Gemini 4 Hydrogen provides four levels of thought — low, medium, high, and extra-high — enabling users to select the reasoning depth according to their needs. Google stated that the model costs USD 1.50 for every million input tokens and USD 7.50 for each million output tokens within its one-million-token context window.

Google's assessments reveal that Hydrogen has increased its score on the Terminal-Bench 4.1 coding benchmark to more than three times that of Gemini 3.8 Flash, achieving 57 percent compared to 18.2 percent. It also decreased false claims by over five times.

Google is also launching Orchestrator in Antigravity, a function that automatically alternates between AI models throughout a session based on the task at hand. On Terminal-Bench 4.0, Orchestrator reached task success rates nearly matching those of Gemini 4 Carbon, while utilizing 24 percent less quota on average and finishing tasks 19 percent more quickly.

Also Read: Uttarakhand Boy Ravi Tamta Builds Electric Flying Car: HAPIDA SKYNeX

The company stated that Hydrogen can operate in conjunction with other models, with one managing coordination while others perform tasks like coding, testing, and verification. It is also intended for research, data examination, and activities involving documents, spreadsheets, graphs, audio, and video.