Two evaluated models meet the same quality threshold for a narrow summarization task. One uses substantially fewer inference resources for the expected workload. Which selection consideration best aligns with responsible resource use?

1. Eliminate quality evaluation because the lowest compute usage always makes a model responsible.
2. Always choose the largest model, regardless of equivalent measured task quality.
3. Generate the longest possible response to every request to maximize useful computation.
4. Consider the more resource-efficient model while confirming it continues to meet quality, reliability, and other business requirements.