A nightly analytics workload processes 5M documents through Claude. There is no user waiting for the result. Which API path minimizes cost?
Answer: Batch API (50% discount, up to 24h delivery)
The Batch API gives a 50% discount in exchange for asynchronous return within 24 hours. Nightly analytics has no user-facing latency requirement and easily tolerates a multi-hour delay, making batch the obvious choice. Streaming does not change cost; parallel sync calls cost full price.