---
title: "Context window: explained simply for businesses"
description: "What is a context window? An AI's working memory: how much text it can take into account at once. Explained simply, with an example and its limits."
url: https://osai.solutions/en/ai-glossary/kontextfenster
lang: en
dateModified: 2026-11-13
---

# Context window

*Also called: Context length, AI working memory*

**In short:** The context window is a language model's working memory: the amount of text it can take into account at once when answering, measured in tokens. Whatever does not fit or drops out, the model does not know at that moment.

## What is a context window?

A [language model](https://osai.solutions/en/ai-glossary/sprachmodell) has two kinds of knowledge. One comes from training and is built in. The other is the **context window**: the text the model has in front of it right now. Anthropic calls it the model's “working memory”.

A good picture is a desk. On it lie your question, the instructions, attached documents and the conversation so far. The desk has a fixed size. Once it is full, nothing more fits, or old sheets have to come off.

## How does it work?

The context window is measured in [tokens](https://osai.solutions/en/ai-glossary/token), meaning word pieces. According to Anthropic, everything that belongs to a request counts:

- the fixed base instruction, the [system prompt](https://osai.solutions/en/ai-glossary/system-prompt)
- the conversation so far, question by question and answer by answer
- attached documents, images and tool results
- the answer the model is writing right now

## Why does it matter for your business?

The size decides how much you can give an AI at once. According to Anthropic, current Claude models such as Opus 5.5, Sonnet 5.5 and Haiku 5.5 have a context window of 1 million tokens, other Claude models 200,000 (as of November 2026).

More is not automatically better, though. According to Anthropic, accuracy and recall drop as more tokens fill the window. Handing the AI the whole filing cabinet makes its job harder than giving it the right pages. Whatever gets lost in the window, the model may replace with a plausible guess ([hallucination](https://osai.solutions/en/ai-glossary/halluzination)).

## How does OsAI use it?

That is why OsAI uses [RAG](https://osai.solutions/en/ai-glossary/rag) for questions about your own documents: the system first finds the matching passages and puts only those into the context window with the question. On the [how it works](https://osai.solutions/en/how-it-works) page you can try this in the demo “Questions to the project file”.

## Frequently asked questions

### Why does ChatGPT forget what I said at the start?

Because long conversations fill the context window, and chat apps may then drop older parts. According to Anthropic, accuracy also drops as the window fills up. Start a new conversation for a new topic and repeat important details.

### Is the context window the AI's memory?

Only its working memory for the current request. What the model learned in training is built into the model. Anything kept between two conversations is stored by the app at most, for example as notes it later puts back into the window.

### Can I give an AI a whole folder?

Often yes: according to Anthropic, a request to current Claude models may contain up to 600 images or PDF pages (as of November 2026). It usually works better to include only the relevant documents, because accuracy drops with volume. With many documents, RAG does the selecting.

## Sources

- [Anthropic: Kontextfenster, Claude-Dokumentation (abgerufen 13.11.2026)](https://platform.claude.com/docs/en/build-with-claude/context-windows)
- [Anthropic: Glossar der Claude-Dokumentation, Eintrag Context window (abgerufen 13.11.2026)](https://platform.claude.com/docs/en/about-claude/glossary)

Last updated: 2026-11-13
