Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Jize, Wu, Han, You, Zhiyuan, Song, Yiming, Wang, Yijun, Shan, Zifei, Li, Yining, Zhang, Songyang, Le, Xinyi, Chen, Cailian, Guan, Xinping, Tao, Dacheng
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2601.18130
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918305151844352
author Wang, Jize
Wu, Han
You, Zhiyuan
Song, Yiming
Wang, Yijun
Shan, Zifei
Li, Yining
Zhang, Songyang
Le, Xinyi
Chen, Cailian
Guan, Xinping
Tao, Dacheng
author_facet Wang, Jize
Wu, Han
You, Zhiyuan
Song, Yiming
Wang, Yijun
Shan, Zifei
Li, Yining
Zhang, Songyang
Le, Xinyi
Chen, Cailian
Guan, Xinping
Tao, Dacheng
contents Mixture-of-Agents (MoA) improves LLM performance through layered collaboration, but its dense topology raises costs and latency. Existing methods employ LLM judges to filter responses, yet still require all models to perform inference before judging, failing to cut costs effectively. They also lack model selection criteria and struggle with large model pools, where full inference is costly and can exceed context limits. To address this, we propose RouteMoA, an efficient mixture-of-agents framework with dynamic routing. It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query, narrowing candidates to a high-potential subset without inference. A mixture of judges then refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference. Finally, a model ranking mechanism selects models by balancing performance, cost, and latency. RouteMoA outperforms MoA across varying tasks and model pool sizes, reducing cost by 89.8% and latency by 63.6% in the large-scale model pool.
format Preprint
id arxiv_https___arxiv_org_abs_2601_18130
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle RouteMoA: Dynamic Routing without Pre-Inference Boosts Efficient Mixture-of-Agents
Wang, Jize
Wu, Han
You, Zhiyuan
Song, Yiming
Wang, Yijun
Shan, Zifei
Li, Yining
Zhang, Songyang
Le, Xinyi
Chen, Cailian
Guan, Xinping
Tao, Dacheng
Artificial Intelligence
Mixture-of-Agents (MoA) improves LLM performance through layered collaboration, but its dense topology raises costs and latency. Existing methods employ LLM judges to filter responses, yet still require all models to perform inference before judging, failing to cut costs effectively. They also lack model selection criteria and struggle with large model pools, where full inference is costly and can exceed context limits. To address this, we propose RouteMoA, an efficient mixture-of-agents framework with dynamic routing. It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query, narrowing candidates to a high-potential subset without inference. A mixture of judges then refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference. Finally, a model ranking mechanism selects models by balancing performance, cost, and latency. RouteMoA outperforms MoA across varying tasks and model pool sizes, reducing cost by 89.8% and latency by 63.6% in the large-scale model pool.
title RouteMoA: Dynamic Routing without Pre-Inference Boosts Efficient Mixture-of-Agents
topic Artificial Intelligence
url https://arxiv.org/abs/2601.18130