Token

Definition: The small chunks of text a model actually reads and writes — not quite words, not quite characters. A token is roughly four characters of English on average, and every model's pricing, context window, and output limits are measured in tokens, not words.

Why it matters: Since cost and limits are counted in tokens, not words, a rough word count can seriously mislead you about what you can actually afford to send or how much a request will cost — especially since Arabic and other non-Latin scripts typically use more tokens per word than English does.

Example: The word "unbelievable" might be split into three tokens ("un", "believ", "able") rather than counted as one — which is part of why token counts run higher than a simple word count.