Tokens
The tiny chunks of text an AI reads, one at a time.
Before an LLM processes text, it breaks it into tokens — small fragments typically representing one word, part of a word, or a punctuation mark. "Unbelievable" might become ["un", "believ", "able"] — three tokens.
Models don't see letters or words; they see token IDs. The entire prompt and response are sequences of these IDs.
Token count matters for two reasons: cost — most APIs bill per 1,000 tokens — and limits — text must fit within the model's context window. Rule of thumb: ~1 token ≈ 0.75 English words. A page of prose ≈ 500 tokens.
In plain terms
Like a musician reading sheet music note-by-note rather than hearing the whole symphony — the model processes token-by-token, building meaning step by step.