What it means
Tokens are the small units of text that a language model reads and writes, often whole words, parts of words or punctuation marks. Models count their input and output in tokens, and providers usually charge by the number of tokens processed.
Real-world example
A ministry estimates the cost of summarising 10,000 case files per month with a cloud AI service. Each file averages 4,000 words, and German texts with long compound words like "Verwaltungsvorschrift" often split into more tokens than equivalent English texts, so the team tests real documents rather than relying on a vendor's English-language estimates.
Why it matters for AI adoption
Tokens determine both cost and limits. They affect how much text fits into a single request and how much a service costs at scale. Budget estimates based on "number of users" rather than expected token volume often turn out to be wrong.
Questions to ask
- How many tokens do we expect per task and per month, and what will that cost at full scale?
- Have we tested token counts with our real documents and languages?