Route easy requests to low-cost models and reserve premium models for complex or high-risk tasks. For large, non-interactive workloads, use Batch inference: submit many prompts as JSONL to Amazon S3 and get responses back asynchronously at a discount versus on-demand โ note it doesn't support tool calling or structured output, since each record is processed independently with no back-and-forth.
Track cost per successful outcome, not cost per request alone.