The cost of computing power and services in the field of artificial intelligence is entering a new stage of refined management. Recently, Amazon has adjusted its pricing plan for the Anthropic model service on its cloud platform, clearly shifting from the current "computing hour" billing to a "token count" based billing model.

This change in pricing logic essentially reflects an upgrade in how cloud computing services account for AI model invocation costs. Although traditional billing methods are simple, they struggle to accurately measure the actual computing power consumption under different task loads. By switching to token-based billing (where tokens refer to the text units processed by the model), Amazon aims to establish an economic model that better aligns with the operational logic of large models. Industry analysts believe that this change in billing model may lead to an increase in operational costs for businesses that frequently call models or handle long texts.

It is reported that this new pricing policy will take effect next year. As generative AI continues to be applied more deeply in the enterprise market, how to balance the return on computing power investment (ROI) has become a challenge faced by tech giants and downstream developers alike. This innovation in Amazon's billing strategy is undoubtedly prompting companies to more carefully optimize their use of tokens during the selection of models and application development stages.