When making model selection decisions, monitoring token spend, optimizing for cost-performance tradeoffs, or when budget limits are approaching.
