A company is building a GenAI application using the Gemini API. They want to minimize latency and cost for a high-volume use case. Which strategy is MOST effective?