How we measure savings
Quick answer
We count characters exactly. Tokens are an estimate from a simple formula, not your provider's tokenizer. We do not see your AI bill, your plan or your credits, so we never claim money saved. To know your real saving, compare your provider's own usage numbers.
Characters: exact
The Platform Lab and the demo on the token page compare the text before and after, character for character. That number is exact.
Tokens: an estimate
The token numbers use the same formula as the Typo app: about one token per four characters, or a count based on words and punctuation when that is higher. Real tokenizers differ per provider and model, so treat the number as an indication. Typo only keeps a change when the estimate goes down, so some changes that would save a few characters are not made at all.
What we do not measure
- Your provider's real token count.
- Earlier messages, files, system instructions, tools and the AI's answer.
- Credits, message limits or what your plan costs.
- Whether the answer you get is as good.
How Typo decides to keep a change
Every rule runs on your text with the protected parts masked: code, links, numbers, quotes, file paths and names. After each rule Typo checks that the protected parts, negations and strength words are unchanged and that not too much was removed. If a check fails, that rule is dropped. If the result is not shorter by the estimate, the original stays.
How to check real savings yourself
- Pick one provider and one model.
- Send the original and the shortened prompt for the same task.
- Read the token usage your provider reports, where it shows it.
- Check that both answers are good enough for the task.
- Count retries. A second message costs more than any saving.
Official sources
Reviewed 8 November 2026. Links go to each provider's own documentation. Typo is not affiliated with any of these companies, and provider terms can change.