How to Count Tokens (and Why Your Estimate Is Usually Wrong)
Every AI bill starts with a token count, yet most people guess it from word counts and get it badly wrong. Here is how tokenization really works and how to get an accurate number before you estimate cost.
A token is not a word
English prose runs roughly four characters — about three-quarters of a word — per token, but that ratio is only an average. Rare words, punctuation and formatting all split into extra tokens, so a clean word-count guess can be off by 20 to 40 percent.
Code and other languages tokenize worse
Source code, JSON and non-English scripts pack far more tokens per character than plain English, because the tokenizer was trained mostly on English text. If your prompts contain code or other languages, count them directly instead of trusting a prose ratio.
Calculators for this topic
Frequently asked questions
How many tokens is a word?+
About 1.3 tokens per English word on average, but paste your real text into the token counter for an exact figure.
Do different models count tokens differently?+
Yes — each model family uses its own tokenizer, so the same text can be a different token count on GPT, Claude or Gemini.