Gemini 4

Positioning Overview

This is a reasoning model for text generation that accepts text and image inputs and uniformly outputs text. Input and output share a context window of up to 1,048,576 tokens, making it suitable for processing long documents and image materials. Aimed at developers who need complex reasoning and large-scale text output, it can be used in everyday production scenarios such as summarization and Q&A, information extraction, report and code writing; it performs in a balanced way on software engineering and knowledge work tasks.

Core Highlights

• Text and image input with unified text output——It can accept text and images, and the same model performs understanding and generates text results, making it easy to process various materials in a unified way.
• Million-level context shared by input and output——The context window is up to 1,048,576 tokens, with long inputs and long outputs sharing the same budget, suitable for processing extremely long materials.
• Single output limit of 65,536 tokens——The maximum output per request is 65,536 tokens, which can support long reports and large-scale code generation.
• Three levels of adjustable reasoning intensity——Use the reasoning_effort parameter to choose among low, medium, and high, trading off between reasoning depth and response speed.

Capability Description

Text and image understanding and generation: After the model understands input text or images, it uniformly outputs text, making it easy to process text materials and image materials through the same interface.

Ultra-long context processing: Input and output share a window of up to 1,048,576 tokens, suitable for holistic understanding and Q&A over long documents and long code.

Controllable long-text generation: The maximum output per request is 65,536 tokens, which can be used to generate long reports, long documents, and large blocks of code text.

Adjustable reasoning intensity: Use reasoning_effort to choose among low, medium, and high, trading off between reasoning depth and response speed according to task complexity.

Everyday tasks oriented toward reasoning: Suitable for complex reasoning and everyday Q&A; after inputting questions and background materials, it outputs text answers, covering most software engineering and knowledge work.

Applicable Scenarios

• Long document summarization and Q&A——Developers input long reports or documents to obtain summaries, Q&A results, and information extraction text for knowledge bases and retrieval systems.
• Image content information extraction——Input screenshots or pictures to obtain the text descriptions and key points within them, for use in material organization and content archiving processes.
• Long report and long code writing——Input requirements and reference materials to generate reports or code text of up to 65,536 tokens, for document delivery and development assistance.
• General complex reasoning Q&A——Oriented toward everyday Q&A and complex reasoning tasks; input questions and background materials and output text answers, for use in assistant-type applications.

Limitations and Selection Advice

The output is text only and cannot generate images, audio, or video. The context window is shared by input and output, so long inputs consume the token budget available for generation, and the maximum output per request is 65,536 tokens. This interface currently does not support tool calling, web search, computer use, or structured output; tasks requiring these capabilities should be considered separately. Reasoning intensity can be selected among low, medium, and high via reasoning_effort.