MetaRoute-Bench：评估智能体工作流路由的元决策策略
原标题：MetaRoute-Bench: Evaluating Meta-Decision Policies for Agentic Workflow Routing
AI 摘要
arXiv 发布论文《MetaRoute-Bench: Evaluating Meta-Decision Policies for Agentic Workflow Routing》，提出一个开放、可检查的基准框架，用于在共享执行模型下比较智能体工作流路由的元决策策略。基准包含 180 个合成任务、8 种路由策略和 30 对随机种子，在 43,200 条轨迹中，任务感知的组合策略成功率达 79.4%，优于静态策略（76.7%）、一次性路由（67.4%）和直接回答（52.9%），但成本增加 4.7%、延迟增加 6.4%。研究发布任务生成、策略、轨迹等工件，主要贡献是可复现的评估方法和路由策略权衡分析。
正文节选
Computer Science > Machine Learning Title:MetaRoute-Bench: Evaluating Meta-Decision Policies for Agentic Workflow Routing View PDF HTML (experimental) Abstract:Agentic systems must repeatedly decide whether to answer directly, decompose a task, invoke a tool, execute code, delegate to a specialist, verify an intermediate result, or recover from failure. These meta-decisions affect not only task success but also operating cost and latency, yet they are often embedded inside an orchest