---
language: en
title: "Context window"
description: "A context window is the token space available for a model’s inputs and response. Its limit affects the documents, history, and instructions used."
canonical: "https://tropk.ai/en/resources/glossary/context-window"
last_modified: "2026-10-12T16:13:35.526Z"
---

# Context window

Reviewed on 2026-10-06.

## What it means

A context window is the amount of information a language model can consider in a single call, measured in tokens. Depending on the implementation, this space includes instructions, messages, documents, tool results, and the generated response. Anthropic’s documentation describes this capacity as working memory, distinct from the dataset used for training.

## Why it matters for marketing

For marketing, the limit determines whether a brief, references, and history fit in the same request. Sending many files without reserving space for the response can interrupt the task. Even when everything fits, the team needs to check whether key rules were applied: temporary storage capacity does not ensure uniform understanding of every passage included in the call.

## Practical example

Imagine an agency requests a catalog review and includes an extensive approval history. The estimated token count exceeds the selected model’s context window before tokens are reserved for the assessment. The agency splits the catalog by product line and keeps the brand rules in each request, preserving the review criteria.

## How to apply or check

Check the capacity of the model and the interface being used, as well as the token count for the complete request. The person responsible for the integration should account for instructions and tool results, reserve space for the output, and watch for signs of truncation or interruption. To assess quality, include a verifiable condition in the document and check whether the response follows it, including with long inputs.

## Important distinctions

A larger context window does not automatically expand the model’s learned knowledge. Nor is it equivalent to the history an interface saves for the user: the application may summarize or remove parts before sending them. Tokens do not correspond directly to words, and input and output limits should be checked for the configuration actually in use.

## Sources and references

- [Context windows](<https://platform.claude.com/docs/en/build-with-claude/context-windows>) — Anthropic

## Continue a leitura

- [Parametric memory](</en/resources/glossary/parametric-memory>)
- [RAG](</en/resources/glossary/rag>)
- [Token](</en/resources/glossary/token>)

## About the author

[Vinicius Pacheco](</en/team/vinicius-pacheco>)

Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content.

---

[All pages in Markdown](<https://tropk.ai/sitemap.md>)