Definition
A token is the basic unit of text that a large language model reads and writes. Before a model processes your prompt, the text is split into tokens: common words become one token, longer or rarer words become several, and punctuation and spaces count too. In English, one token is roughly three quarters of a word, so 1,000 tokens is about 750 words. Models generate their answers one token at a time.
For a business, tokens matter for two reasons: cost and limits. Most AI APIs charge per million tokens, with separate prices for input (what you send) and output (what the model writes). A long document pasted into every request multiplies your bill. Tokens also define the context window, the maximum amount of text a model can handle at once. Engineers design prompts and retrieval so that only relevant text consumes tokens.
Token prices have fallen sharply year after year, and techniques such as prompt caching cut the cost of repeated context further. Still, an agent that runs hundreds of steps can consume millions of tokens a day, so monitoring matters. A common misconception is that a token equals a word or a character. It is neither, and languages such as Spanish or German often use more tokens per sentence than English.
In practice
A company summarising 10,000 customer calls a month sends about 3,000 tokens per call and gets 200 back. That is 32 million tokens a month, a number the finance team can now price and forecast like any other utility.
Why it matters
Tokens are the unit of AI cost. If you approve AI budgets, you should know roughly how many tokens your use cases burn and what each one is worth.
Frequently asked questions
- How many tokens are in a word?
- In English, a token is roughly three quarters of a word on average, so 100 tokens is about 75 words. Short common words are usually one token; long or technical words are split into several. Other languages and code typically need more tokens for the same content.
- Why do AI companies charge per token?
- Because computing cost scales with the amount of text processed and generated. Input tokens are cheaper than output tokens since the model reads faster than it writes. Per-token pricing lets you estimate cost from volume, much like paying per message or per minute.