You are deploying a model in Azure AI Foundry and must ensure the endpoint can...

Choose a Standard pay-as-you-go deployment . Microsoft Foundry documentation identifies Standard deployments as appropriate for low-to-medium volume workloads with high burstiness . Standard uses pay-per-token billing , so charges are based on actual inference consumption rather than continuously reserved model capacity. This makes it well suited to workloads whose traffic varies significantly and may have long idle periods.

By contrast, Provisioned Throughput Units (PTUs) reserve model-processing capacity and are designed for sustained, predictable high-throughput workloads where consistent latency is important. Microsoft states that PTUs are billed based on deployed capacity even when that capacity is not actively serving requests. Consequently, PTU deployments can be economically inefficient when the endpoint is frequently idle or traffic is highly variable.

A Batch deployment is optimized for asynchronous, non-latency-sensitive processing and is not appropriate for an interactive endpoint receiving unpredictable live traffic. Dedicated managed compute with fixed capacity introduces continuously allocated infrastructure and therefore does not align with the requirement to minimize idle-period cost.

Microsoft also recommends Standard or Global Standard for variable, bursty traffic , while provisioned types are recommended for consistent high-volume workloads.

Study Guide references: Microsoft Foundry Models → Standard deployments; pay-per-token billing; bursty workloads; provisioned throughput; batch processing.