Abstract 3D scene in 21:9 with cool-toned glass knowledge overlays and light trails; a stack of translucent glass panels on the right reads gemini-4, soft blank light on the left, gradient of cream, ice blue, and coral, serving as the hero image for the gemini-4-argon detail page.

Google

Gemini 4

gemini-4-hydrogen

Google Gemini 4 text reasoning series: Carbon is the most powerful in the series, aimed at complex, multi-step, long-horizon tasks; Argon offers a better balance of capability and cost. Supports text and image input, 1M token context

Route
Context window
1,048,576
Input
Max output
65,536 tokens
Release date
2026-11-12
Reasoning Streaming

Try gemini-4-hydrogen online

Pricing

gemini-4-hydrogen /1M tokens

Specifications Unit Price
Input $2.26757
Cached input $0.11338
Output $11.33787

Overview

Positioning overview

This is a text-output model aimed at developers and professional teams that need long-context processing. It accepts text, image, audio, video, and file inputs and consolidates them into text results, making it suitable for information synthesis and content generation across multiple materials. Everyday coding, debugging, and multi-step agent workflows are its main focus, and it can also handle long-document understanding, multimodal material synthesis, and professional writing tasks that require sustained context.

Key highlights

  • Unified context window of up to 1 million tokens — can hold long documents and multi-step context within a single task, reducing the need for mid-way segmentation and reorganization.
  • Default synchronous output of up to 64k tokens — regular requests return text directly, and long outputs can be extended via background long decoding.
  • Three adjustable thinking levels (low / medium / high) — trade off response speed against reasoning depth according to task needs, adapting to different development stages.

Capability description

Multimodal input to text: accepts text, image, audio, video, and file inputs, consolidating mixed materials into one or more text results for subsequent retrieval and processing.

Long-context handling: a unified context window of up to 1 million tokens, suitable for keeping long documents, codebase documentation, and multi-turn conversations in the same task for continuous processing.

Long-output continuation: default synchronous output of up to 64k tokens; when longer text is needed, background long decoding and continuation can extend the output to up to 1 million tokens.

Thinking level adjustment: offers three configurable thinking levels (low / medium / high), letting developers balance response speed and reasoning depth for coding, debugging, or synthesis tasks.

Applicable scenarios

  • Developers performing repository-level code processing — input code files and requirement descriptions to get modification suggestions, debugging ideas, or interactive web development snippets for everyday coding work.
  • Building multi-step agent workflows — place tool-call results and intermediate states into the context to continuously obtain the next step's text instructions, suitable for automated task orchestration.
  • Document understanding across mixed materials — consolidate text, tables, charts, and image materials to output a unified text summary or analysis for report writing and knowledge organization.
  • Professional writing assistance for long documents — combine long-context processing of research materials and drafts to generate structured text content for first drafts of professional documents in finance, legal, and other fields.

Limitations and selection advice

The output type is text; it does not directly generate images, audio, or video. The default synchronous output limit is 64k tokens, and reaching the maximum 1 million output tokens requires background long decoding or continuation. With multimodal prompts, improper refusals are relatively more noticeable, so a manual review step should be reserved when sensitive content is involved.

  • New model New Models Available: Gemini 4 Carbon, Gemini 4 Argon, Gemini 3.8 Flash

Related Models

View all models →

Same Type