Cost Optimization

Prefer -mini / -nano / -flash models for simple tasks
Use a larger-context model for long documents, a cheaper one for short tasks
Set a budget cap to prevent overspending
Reuse prompt context and enable caching in batch workloads