---
title: "Token in AI: explained simply for businesses"
description: "What is a token in AI? A word piece that language models compute with and providers bill by. Explained simply, with an example and what it means for costs."
url: https://osai.solutions/en/ai-glossary/token
lang: en
dateModified: 2026-11-13
---

# Token

*Also called: Tokens, Word piece*

**In short:** A token is a building block of text, usually a word or part of a word, into which a language model splits everything it reads. The model computes in tokens, and providers bill in tokens, separately for input and output.

## What is a token?

A [language model](https://osai.solutions/en/ai-glossary/sprachmodell) does not read whole sentences. It splits every text into **tokens**. According to Anthropic, a token can be a word, part of a word, a single character or even a byte. Short, common words often stay whole; a long German word like “Dachrinnenreinigung” (gutter cleaning) is usually split into several pieces.

Think of Lego bricks: the model builds every text from bricks, and what gets counted is the bricks, not the words.

## How does it work?

According to Anthropic, how many characters a token covers depends on the model and the language. Newer Claude models produce around 30 % more tokens for the same text than older ones. So there is no fixed conversion; only a token count with the model gives the exact number.

Tokens arise in two places:

- **Input:** everything the model has to read, meaning your [prompt](https://osai.solutions/en/ai-glossary/prompt), background instructions, attached documents and the conversation so far.
- **Output:** every word piece the model writes as its answer.

## Why does it matter for your business?

Tokens are the unit behind the costs. Anthropic bills use of its interface ([API](https://osai.solutions/en/ai-glossary/api)) per million tokens, and output costs more per token than input (price list, November 2026). A workflow that reads many emails a day keeps using tokens, even when nobody watches.

Tokens also limit how much a model can handle at once. That upper limit is called the [context window](https://osai.solutions/en/ai-glossary/kontextfenster).

## How does OsAI use it?

In the workflows OsAI builds, a tool such as n8n handles the fixed steps and the AI only handles understanding text. So the model reads one enquiry instead of the whole inbox, and the relevant passage instead of the whole file. That saves tokens.

The German guide [on AI automation costs](https://osai.solutions/de/blog/ki-automatisierung-kosten) (in German) covers running costs. The [AI ROI calculator](https://osai.solutions/en/ai-roi-calculator) shows whether a workflow pays off for you.

## Frequently asked questions

### How many words are in a token?

There is no general answer: according to Anthropic, it depends on the model and the language. Short, common words are often one token, long or rare words several. A token count with the model you use gives the exact number.

### Is AI billed by tokens?

Through the API, yes: Anthropic, for example, bills per million input and output tokens. On a chat subscription with a fixed monthly price you do not pay per token, but tokens show up as usage limits. Automated workflows usually run through the API.

### What is the difference between a token and a context window?

The token is the unit; the context window is the upper limit: how many tokens a model can take into account at once. For current Claude models, according to Anthropic, that is up to 1 million tokens (as of November 2026).

## Sources

- [Anthropic: Glossar der Claude-Dokumentation, Eintrag Tokens (abgerufen 13.11.2026)](https://platform.claude.com/docs/en/about-claude/glossary)
- [Anthropic: Preise der Claude-API, Abrechnung nach Eingabe- und Ausgabe-Tokens (abgerufen 13.11.2026)](https://platform.claude.com/docs/en/about-claude/pricing)

Last updated: 2026-11-13
