Gemini 3.8

Positioning Overview

This is a fast, low-cost general-purpose model built for text output. It accepts text and image inputs and returns text, with a 1,048,576-token context window and up to 65,536 tokens of output per call. It is suited to high-concurrency, low-latency everyday text tasks, as well as agentic coding, enterprise knowledge-work automation, and multi-step reasoning in specialized domains; developers can use it for document comprehension, summary generation, and step-by-step conclusions.

Core Highlights

• Million-scale context with 64k output — a 1,048,576-token context window paired with up to 65,536 tokens of output per call, covering long-range, large-input tasks.
• Text and image inputs — it directly accepts text and images and uniformly returns text results, making it easy to handle mixed text-and-image materials consistently.
• Adjustable reasoning depth — reasoning_effort supports three levels: low, medium, and high, letting you balance reasoning investment against responsiveness according to task complexity.

Capabilities

Text and image comprehension: Given text or images, it can understand text-and-image content, summarize it, and answer questions, with the final result in text — well suited to document review and knowledge organization.

Long context and long output: It supports a 1,048,576-token context and outputs up to 65,536 tokens per call, suited to understanding long documents and to long-range task chains.

Adjustable reasoning intensity: reasoning_effort can be set to low, medium, or high, adjusting reasoning investment to task complexity; minimal is not supported.

Use Cases

• Agentic coding for developers — input code and requirement text to get fix suggestions or implementation plans, automating progress on long-range software engineering tasks.
• Enterprise knowledge-work automation — input business documents and process descriptions to produce organized text conclusions, supporting decision-making and execution in complex workflows.
• Text-and-image document summarization — input long documents or image materials to generate text summaries and key points for document review and knowledge archiving.
• Multi-step reasoning in specialized domains — input professional materials in fields such as finance and law to produce step-by-step reasoning and conclusions in text for agentic analysis tasks.

Limitations and Selection Advice

Output is text only; it does not perform speech synthesis or real-time voice conversation. Inputs support only text and images. The thinking level does not support minimal and can only be set to low, medium, or high. The per-call output limit is 65,536 tokens, so overly long content must be processed in multiple passes.