Token

How does Token work?

1. 1Before reading, the AI cuts text into small pieces called tokens: sometimes a whole word, sometimes part of a word or a punctuation mark.
2. 2The prompt, attached documents and chat history all count as input tokens.
3. 3The answer the AI writes counts as output tokens, usually priced higher than input.
4. 4AI prices and the limit of each session (the context window) are both measured in tokens.

What a shop or café can use it for

• Estimating chatbot costs: messages per month times average tokens per exchange.
• Seeing why pasting the whole catalogue into every question makes the AI bill jump.
• Choosing an AI plan that matches real use rather than the biggest one.

What to watch out for

• Vietnamese with accents often uses more tokens than English for the same idea.
• The longer the conversation, the more each turn costs, since the AI rereads the history.
• When comparing services, compare both input and output token prices.

Quick questions

• How many words is a token?

  It varies by language and model. In English a token is usually shorter than a word; Vietnamese with accents tends to need more tokens for the same number of words.

• How can I use fewer tokens?

  Send only relevant passages (RAG), keep prompts tight, start a new chat when the topic changes, and cap the answer length.

• Do I need to worry about tokens with a ChatGPT subscription?

  Consumer subscriptions usually do not charge per token but do limit usage. Tokens matter when you pay through an API, for example for a shop chatbot.