Token counter
Estimate how many tokens your text uses in AI models. The calculation runs in your browser: we never send your text to a server.
The figure is an estimate (~3.75 characters per token). Each model uses its own tokenizer, so the exact number may vary slightly. To work out the cost, use our API cost calculator.
What a token is
Language models do not read letters or words: they split text into chunks called tokens. A token is usually a whole short word, a fragment of a long one, a punctuation mark or a space. "Hello" is one token; "intelligence" splits into several. As a rough rule, a token is about 3.75 characters, so a page of text runs to 600-700 tokens and a 200-page book approaches half a million.
Why the token count matters
For two separate reasons. The first is money: if you use the API you pay per input token and per output token, so the number is literally your bill. The second is the limit: every model handles a maximum number of tokens across what you send and what it replies, and past that the answer gets cut off or it forgets the start of the conversation. If an assistant has ever lost the thread of a long document, that is why.
Spanish uses more tokens than English
The tokenizers behind the main models were trained mostly on English text, so English words split into fewer chunks. The same content in Spanish uses roughly 20-30 % more tokens, and in Catalan, Basque or Galician the gap is wider. That has two practical consequences: a Spanish-language service costs more to run than the same one in English, and you hit the context limit sooner.
How to use fewer tokens
- Do not resend the whole conversation on every request if you do not have to: summarise the older turns and send the summary.
- Trim repeated instructions. A 500-token system prompt that rides along on every call is paid for on every call.
- Ask for replies the length you actually need. Output is the expensive part, and "be brief" is an instruction that translates into money.
- If you send documents, send the relevant fragment rather than the whole file.
How accurate this estimate is
The calculation runs in your browser using an average characters-per-token figure, so it is an approximation, not the exact number your provider will bill. Each model family uses its own tokenizer and the usual deviation is a small percentage — enough to budget with, not to reconcile an invoice. For that, each provider publishes its own official counter. Nothing you paste here leaves your machine: we do not send the text to any server.