Tokens generated by the model.
Output tokens are the tokens the model writes back, including reasoning tokens when a provider bills them. Output rates are typically several times the input rate.
Long answers, chain-of-thought reasoning, and translation workloads are output-heavy and sensitive to output pricing.