Your assistant uses a classifier to route queries to a small or large model, which saves 60% of GPU spend. Analysis shows about 20% of hard queries are misrouted to the small model, which then gives confident wrong answers, and regenerate clicks on those threads have doubled. What should you do?

1. A Send all traffic to the large model, since misrouting costs more in lost user trust than the GPU savings could justify at any scale.
2. B Route on prompt length alone, since long prompts correlate with difficulty and length is free to compute before inference.
3. C Bias the router toward the large model when uncertain and retrain it on regenerate and thumbs-down signals, accepting some cost.
4. D Cascade every query: the small model answers and the large model verifies each response before display as a quality gate.