A base model hosted on Amazon Bedrock is being prompted with 10 examples once per day and is performing well. The company wants to reduce monthly costs. Which action will lower costs while keeping the same base model?

Reducing the number of tokens in each prompt directly lowers costs because Amazon Bedrock charges for inference based on the number of input and output tokens processed. Fewer tokens mean lower processing costs for each daily interaction. Customizing the model by fine-tuning it would incur additional costs for training and hosting the custom model, increasing overall expenses rather than reducing them. Increasing the number of tokens in each prompt would lead to higher costs, as more tokens are processed per request. Using Provisioned Throughput offers dedicated capacity and predictable performance but typically comes at a higher cost than on-demand pricing, especially for a low-volume use case like 10 prompts per day.