A team routes all tasks to its largest model. Many requests are simple classifications, while a small subset requires difficult reasoning. Which optimization should be evaluated?

1. Route using only the user name length.
2. Randomly route models and omit version tracking.
3. Evaluate quality-aware routing to suitable models with measured fallback behavior.
4. Route every request to the cheapest model without checking task success.