A media company runs two Claude workloads on the same large model: a real-time newsroom assistant, and a nightly job that tags and summarizes about 60,000 archived articles. The nightly job accounts for most of the spend, and its results are needed by 9 a.m. Quality on the tagging eval is currently acceptable. Which change most directly cuts cost without risking the quality bar?

Answer: D — Move the nightly job to the Message Batches API, keeping model and prompt unchanged

(d) is right: the nightly job isn't latency-sensitive, and the Message Batches API is designed for asynchronous bulk work at lower cost, so moving it there reduces spend without changing what the model sees. (a) downsizes without evidence and risks both workloads. (b) cuts context the tags and summaries may depend on. (c) finishes faster but doesn't reduce cost per request.