Reconceptualizes LLM routing as a MaxSAT constraint optimization problem, where natural language feedback acts as hard and soft constraints.
March 17, 2026
Original Paper
LLM Routing as Reasoning: A MaxSAT View
arXiv · 2603.13612
The Takeaway
Rather than treating model selection as a simple classification or ranking task, this framework allows for mathematically rigorous routing based on partially observable model attributes and complex user preferences. It provides a path toward interpretable and verifiable model orchestration in multi-LLM systems.
From the abstract
Routing a query through an appropriate LLM is challenging, particularly when user preferences are expressed in natural language and model attributes are only partially observable. We propose a constraint-based interpretation of language-conditioned LLM routing, formulating it as a weighted MaxSAT/MaxSMT problem in which natural language feedback induces hard and soft constraints over model attributes. Under this view, routing corresponds to selecting models that approximately maximize satisfacti