An internal tool receives a few hundred requests daily, with many hours of no traffic. Each request performs two seconds of stateless processing, fits within Lambda limits, and can tolerate normal cold-start latency. Which compute design is generally the most cost-effective of the following for avoiding idle server charges?

1. Use API Gateway and Lambda with on-demand execution and no provisioned concurrency.
2. Run two large On-Demand EC2 instances continuously for the tool's peak possible traffic.
3. Buy a three-year dedicated compute commitment sized to ten times the peak load.
4. Keep a Fargate service at ten running tasks even when the tool is idle.