Select-then-Solve: Paradigm Routing as Inference-Time Optimization for LLM Agents
Published:
Authors: Heng Zhou, Zelin Tan, Zhemeng Zhang, Yutao Fan, Yibing Lin, Li Kang, Xiufeng Song, Rui Li, Songtao Huang, Ao Yu, Yuchen Fan, Yanxu Chen, Kaixin Xu, Xiaohong Liu, Yiran Qin, Philip Torr, Chen Zhang, and Zhenfei Yin.
This work shows that no single inference-time reasoning paradigm is best across tasks. A lightweight learned router selects among six paradigms on a per-task basis, improving average performance over the best fixed strategy.