Fine-tune the Bedrock model with domain data so prompts can be shorter while maintaining quality is the correct option.

By applying fine-tuning you align the foundation model to your domain language and tasks so prompts no longer need long context to elicit accurate responses. This reduces both prompt and response tokens used during inference and directly lowers monthly token spend while preserving answer quality.

Raise the model's temperature to produce outputs faster is incorrect because temperature controls output randomness and diversity rather than latency or token usage and increasing it can harm accuracy.

Perform continued pretraining on Amazon SageMaker using the company's full dataset is incorrect because continued pretraining is resource intensive and expensive and it is unnecessary when managed Bedrock models can be customized more cheaply through fine-tuning or instruction tuning.

Lower the maximum tokens per response to reduce output size is incorrect because hard caps can truncate useful content and degrade answer quality which defeats the goal of maintaining high quality even though costs may drop.

Full AWS Practitioner Certification Question