Data
Token
What is token?
A token is the atomic unit a language model reads and writes — typically a common word, a word fragment, or a punctuation mark. Model context limits, throughput and per-token pricing are all denominated in tokens rather than words or characters.
The mental conversion for English prose is about three-quarters of a word per token, so a thousand tokens is roughly seven hundred and fifty words. That approximation fails badly on code and on languages that do not use spaces.
Output tokens are generated one at a time and are therefore the component of latency you feel. Input tokens are processed in parallel and are correspondingly cheaper and faster per token.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.