Token burn refers to the rate at which AI systems consume tokens during input and output processing. Tokens are the units AI models use to read and generate text. A token may be a word, part of a word, punctuation, or a symbol. When you send a prompt to an AI model, the system processes input tokens and then creates output tokens in response.