Question

Your team runs a customer-facing chatbot on Azure AI Foundry backed by a GPT-4o deployment using provisioned throughput units (PTUs). Traffic is steady and predictable at high volume 24/7, and finance wants to reduce the monthly PTU spend as much as possible without changing the deployed model or its throughput guarantees. What should you do?

Accepted Answer

Purchase a monthly or yearly PTU reservation for the committed capacity to obtain a discounted hourly rate.

Answer

Migrate the deployment to a serverless (pay-as-you-go) endpoint to eliminate the fixed PTU cost.

Answer

Reduce the number of provisioned throughput units below the required baseline to lower the hourly charge.

Answer

Enable content safety filtering to reduce the number of billable tokens processed by the model.