✦
Summary
Gemini 4 Hydrogen is Google's fast, efficient tier of the Gemini 4 family, released in November 2026 for everyday coding and agentic workflows. It is a multimodal large language model that takes text, image, video, audio and PDF input and returns text, and it is built to run agents that call tools, execute code and work through multi-step tasks. Google positions it as the quick tier of the family: it scores 51 on the Artificial Analysis Intelligence Index, ahead of Gemini 4 Argon at 50, while running faster, and it is a major upgrade over Gemini 3.8 Flash at about 2.5 times the speed and 3 times the Terminal-Bench 4.1 score.
Key Features
1,048,576-token unified context window for input and output
Four thinking levels: low, medium, high by default, and new xhigh
Multimodal input with text, image, video, audio and PDF, and text output
Function calling, code execution and computer use supported
Grounding with Google Search and Google Maps
Structured outputs and URL context supported
Up to 1M output tokens with long decode, 65,536 tokens by default in synchronous calls
Flat pricing with no long-context surcharge and implicit context caching
About Gemini 4 Hydrogen
The model offers a 1,048,576-token unified context window for input and output, with up to 1M output tokens and four thinking levels: low, medium, high as the default, and a new xhigh level. It supports function calling, code execution, computer use, grounding with Google Search and Google Maps, structured outputs and URL context, with flat pricing and no long-context surcharge plus implicit context caching. Developers reach it through the Gemini API and Google AI Studio with an API key, through the Gemini app, Google Search, Vertex AI and Antigravity.
Gemini 4 Hydrogen is free to try on the Gemini API and AI Studio free tier, with paid use at $1.50 per 1M input tokens and $7.50 per 1M output tokens, flat across the full context. It is a distinct model from Gemini 4 Carbon, the other Gemini 4 tier already listed in the directory, and is aimed at developers building coding assistants and agents that need speed more than maximum reasoning depth.