What Is a Token? AI Models, APIs & Context Windows
Last updated: July 2026
Tokens are how LLMs meter text—not pages or words. Learn what they are, how they differ from word count, and when to use a browser token counter before you paste.
Large language models do not meter text by pages or English words. They split strings into tokens — chunks that may be a whole word, part of a word, punctuation, a space, or a piece of code. Two prompts that look similar in an editor can differ a lot once tokenized.
Words, characters, and tokens
Characters are raw length. Words on this site are a whitespace split only — not Chinese word segmentation — so continuous Chinese can show a low word count while characters stay high. Tokens are what most APIs and context windows use. Dense code, mixed languages, and odd Unicode often change the ratio between all three.
Why bills and windows use tokens
Providers price and truncate by tokens because that is how the model’s vocabulary works under the hood. If you only watch characters, a long paste can surprise you when the request fails or the model “forgets” earlier turns.
When a counter helps
Use the token counter before you hit a context limit, stuff a ChatGPT composer, or spend budget on an oversized API call. Prefer OpenAI-style (tiktoken) modes for GPT-family length; treat Claude and Gemini as labeled estimates. For a word-count brief, try words to tokens. For fill %, use the context window calculator.
How this relates to “online token counter” searches
Queries like AI token counter, LLM token counter, and online token counter usually want the same job: paste and read length. One workbench covers those intents better than a stack of near-duplicate pages. Vendor-specific nuance lives in the guides below.