Learn
What is a token in AI?
Last reviewed July 16, 2026
A token is the unit of text an AI model actually reads and writes: a word, part of a word, or a punctuation mark, averaging roughly three-quarters of an English word. Everything about a model is measured in tokens: its context window is a token limit, its price is quoted per million tokens in and out, and its speed is tokens per second. A 1,000-word document is roughly 1,300 tokens.
Why prices are per million tokens
Providers bill input tokens (everything you send: the message, the history, attached files) and output tokens (what the model writes) at separate per-million rates, with output typically several times the input rate. Long conversations resend their history on every message, so the same question costs more in a long thread than in a fresh one. That makes token counts the practical unit of cost control.
Tokens are not words
Tokenizers split text into subword pieces, so counts vary by content and by model family: code, non-English languages, and unusual formatting use more tokens per word than plain English prose. The only reliable way to know a prompt's size is to count it with the model's own tokenizer rather than estimating from word count.
idapt shows a cost estimate before a message sends, prices every catalog model per million tokens on its model page, and counts tokens for any prompt in the token counter tool.
Frequently asked
See it in practice
The fastest way to understand it is to use it. Start free, no credit card.
Try idapt free