A team uses Opus for all calls in their production app. 60% of their calls are simple classification tasks. What is the highest-impact single optimization?
Answer: Route the classification calls to Haiku, keep Opus for complex calls
Model routing — sending classification to Haiku while keeping Opus for complex calls — captures roughly a 10x cost reduction on 60% of the traffic. Sonnet is cheaper than Opus but still 2-3x Haiku for tasks Haiku handles fluently. Batch API only applies to non-interactive workloads. Streaming doesn't change cost.