AI Token Counter

What this tool is for

Tokens are the unit models actually read, charge for, and run out of. They are not words and not characters: roughly four characters or three-quarters of a word in English, and considerably worse in most other languages, where the same meaning can cost twice the tokens.

Knowing the count matters for two practical reasons — whether your input fits, and what it costs.

How it works

Paste text and the counter estimates tokens for the major model families, alongside character and word counts. It blends the standard four-characters-per-token approximation with a word-based estimate, which lands within roughly 10–15% for ordinary English prose.

Everything runs in your browser. Nothing is transmitted, which is the point — a token counter that uploaded your document to count it would be a strange trade.

When to use it — and when not to

Use it when planning a prompt against a context limit, when estimating what a job will cost before you run it at volume, and when deciding whether a document needs chunking before a model sees it.

Do not use it where the exact number matters — billing reconciliation, or a hard limit you are sitting right on the edge of. For those, use the provider's official tokenizer. This is a planning instrument, and an estimate that is 15% out is fine for planning and not fine for accounting.

Things worth knowing

Two things surprise people. Code tokenises far worse than prose — punctuation, indentation and identifiers all cost, so a code file is often double what the character count suggests. And non-English text can cost two to three times more for the same meaning, which quietly makes multilingual applications more expensive than their English equivalents.