Characters, tokens and what you pay

You are billed per character. One character of source text is one billed character — whitespace and markup included. Nothing else enters the price: not tokens, not requests, not the language pair, not the time of day.

Then why do you show tokens?

Because they are the honest measure of the work behind a translation, and because you should be able to see it. Every response from POST /v1/translate carries:

{
  "translations": [{ "text": "Hello world", "detected_source_language": "DE" }],
  "characters": 10,
  "engine": { "prompt_tokens": 812, "completion_tokens": 640, "total_tokens": 1452 }
}

A token is the unit the translation engine reads and writes in: roughly a short word or a piece of one. prompt_tokens is what we sent (your text, the glossary entries that apply, and the instructions that keep formatting intact). completion_tokens is what came back.

Why the number moves for the same character count

• Language. Latin scripts pack more characters into a token than Greek, Cyrillic or CJK.
• Glossaries. Every applicable entry travels with the request.
• Tags and formatting. tag_handling keeps markup intact, and markup costs tokens.
• Segmentation. Long inputs are split; each part carries its own instructions.

None of this reaches your price. Two requests with the same character count cost the same, whether one used three times the tokens of the other.

Where to find them

Cached requests report zero

A request answered from the translation cache never reaches the engine, so it reports total_tokens: 0. The characters are still counted, because the cache is a service we run for you, not a discount we hide. See Usage and quota.

What we do not publish

We do not publish what a token costs us. That is a supplier term, and our price to you does not depend on it — see Pricing.