“The price per token for an AI model that is two generations old is approximately 1/100th of the price of a current-generation model, indicating a rapid cost decline.”