p28 / Preprint
SeLMRoute: Probabilistic Semantic Evidence for Large Language Model Routing
arXiv:2609.34736 · 2026
In plain language
SeLMRoute selects an LLM using explicit probabilistic evidence about what a query requires.
Research context
The separation supports interpretable routing and changing performance or cost priorities without redefining query evidence.
Approach
A decision model answers interpretable semantic probes. A lightweight supervised router estimates candidate performance, with deployment objectives applied afterward.
Findings
On 11,481 queries and 20 models, grouped out-of-fold accuracy is 72.64%, compared with 69.23% for the strongest fixed candidate. A separate 13-model cost setting improves performance across all five splits.
Summary based on full paper.