The chunk of text a model actually reads and bills for.
Models do not see words, they see tokens: fragments of roughly four characters in English. "Leverage" might be one token; an unusual name might be three.
Everything is denominated in tokens — context limits, throughput, and price. A rough conversion is 750 words per 1,000 tokens.
Input and output tokens are usually priced differently, with output several times more expensive because it is generated one step at a time.