🛰️ Daily AI Frontier
‹ back to 2026-08-13

SCOPE-Router: Cost-Aware Open-Set VLM Routing for Execution-Oriented Tasks

arXiv cs.CV Efficiency & Systems Tao Yu, Yifei Qu, Zhiqing Cui, Pengfei Zhou, Zhongtian Luo, Yujia Yang, Shenghua Chai, Haopeng Jin, Zhenghao Zhang, Xinming Wang, Hongzhu Yi, Wangbo Zhao, Zhenglin Wan, Yan Huang, Yeshani, Jinwen Luo, Yang You 2026-08-12
Representative image for SCOPE-Router: Cost-Aware Open-Set VLM Routing for Execution-Oriented Tasks

TL;DR - SCOPE-Router selects VLMs for execution-oriented queries while balancing quality and cost, including when unseen models or tasks appear. It outperforms competing routers in open-set and out-of-distribution evaluations.

  • VLM-ExecRouterBench covers Code, Agentic, and Search tasks across 11 models with widely varying prices.
  • The dual-tower router matches queries against calibrated model behavior profiles, allowing new models without retraining.
  • Its CRM+RCCR objective preserves multiple suitable-model signals while incorporating cost preferences.
  • SCOPE-Router leads all three benchmarks; CRM+RCCR also improves four other router architectures by 1.25–6.21 Rank Score points.

view merged work →