A nightly account-research agent's cost has doubled. A token breakdown shows most input tokens are raw JSON payloads returned by a CRM tool, of which the agent uses a handful of fields. Results are needed by the next morning, not in real time. Which two changes most directly reduce cost? Select two.

Answer: A and B — Have the CRM tool return only the fields the agent actually uses instead of the full payload / Submit the nightly runs through the Message Batches API, since they aren't latency sensitive

(a) and (b) are right: trimming tool results at the source removes the biggest block of wasted input tokens, and batch processing trades turnaround time the job doesn't need for lower cost. (c) raises per-token cost on a guess that it will reduce calls. (d) doesn't address input bloat, and a higher ceiling doesn't reduce what the model generates when nothing is being truncated.