Glossary
Tokens
The units LLMs use to process text — roughly 3-4 characters or 0.75 words each.
Definition
A token is the smallest unit an LLM processes. Most modern models use subword tokenization: common words are one token, rare or long words split into multiple. As a rough rule of thumb, 1,000 tokens ≈ 750 English words.
Context
Token count is what LLM APIs price on and what context windows are measured in. When budgeting an AI system, cost is dominated by tokens consumed (both input and output). Efficient prompt engineering and retrieval design both reduce token usage substantially. Non-English content typically uses more tokens per word than English.