Your organization needs to process 100,000 short product descriptions daily for sentiment classification into 5 categories. Each description averages 50 tokens. The classification requires Claude's response to be a single category label (1-3 tokens). What API design pattern minimizes cost and latency for this workload?

A is correct because the Anthropic Batches API is purpose-built for exactly this use case: high-volume, non-real-time classification workloads. The Batches API provides a 50% cost discount compared to synchronous API calls, handles the scheduling and rate limiting automatically, and returns results asynchronously — ideal for a daily batch job with no real-time latency requirement. For 100K requests at 50 input + 3 output tokens, the cost savings are substantial. B is wrong because stream cancellation after the first token is a brittle approach — billing for input tokens is not affected by stream cancellation, and the operational complexity of implementing early stream termination at 100K requests/day is unjustified. More importantly, there is no guarantee Claude produces the label as its first tokens. C is wrong because sequential single-thread processing of 100K requests would take many hours even at high API throughput and creates unnecessary rate limit pressure. D is wrong because batching 50 descriptions per request in a JSON array creates response parsing complexity, risks output format confusion (descriptions may interfere with each other's classification), and does not provide the 50% cost discount that the Batches API offers.