Shuhan Wang

SWE-Router: Routing in Multi-turn Agentic Software Engineering Tasks

This work introduces SWE-Router, a value-based temporal approach that lets a cheap model run for a few exploratory turns and reads the resulting partial trajectory before deciding whether to continue cheaply or to escalate to an expensive model, and provides a Bayes-optimality theorem showing that conditioning on the partial trajectory never harms routing and is strictly better whenever exploration is informative.