RetrievalRouter: 2.5% More Accurate and 12.4x Faster via Dynamic Routing

_reachsumit · x · 2026-08-27

The paper introduces RetrievalRouter, a lightweight query-aware router that dynamically selects the best retrieval pipeline (modality and architecture) for each query. Across benchmarks in financial and scientific corpora, it achieves 2.5% higher nDCG@5 and is 12.4 times faster than the best static baseline, resolving the accuracy-latency trade-off.

Related event: RetrievalRouter adaptively selects retrieval pipelines for better accuracy and speed(2 posts)→

Original post →

More from Research

Research channel →