TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AI

ID: 38df02ae
原文链接: https://arxiv.org/abs/2607.22465
PDF: https://arxiv.org/pdf/2607.22465v2
作者: Ritik Raj, Souvik Kundu, Sarbartha Banerjee, Dheemanth Joshi, Ishita Vohra, Tushar Krishna
日期: 2026-07-24
更新: 2026-07-27
分类: agents
来源类型: article
标签: model-routing, contextual-bandit, cost-quality, agentic-workflows, deployment
质量评分: 4/5
抓取时间: 2026-07-29T06:19:39Z

中文导读

TRACE-Router解决了Agent工作流中每次调用独立路由与任务级结果之间的不匹配问题它在任务准入时通过上下文赌博机一次性分配模型，将后续所有调用固定到同一后端，并使用终端奖励联合考虑准确率和延迟更新策略在tau2-Bench上高出7-8个准确率点；在Terminal-Bench上高出最强单模型7.1点且延迟降低36%

为什么值得关注

TRACE-Router解决了Agent工作流中每次调用独立路由与任务级结果之间的不匹配问题它在任务准入时通过上下文赌博机一次性分配模型，将后续所有调用固定到同一后端，并使用终端奖励联合考虑准确率和延迟更新策略在tau2-Bench上高出7-8个准确率点...

本文以 arXiv 摘要与条目既有摘要为据：程序性技能/自进化/形式化研究/路由等主题直接对应 Agent 工程与评估实践中的可靠性、可验证性与部署权衡。

关键信息

论文标题：TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AI
作者：Ritik Raj, Souvik Kundu, Sarbartha Banerjee, Dheemanth Joshi, Ishita Vohra, Tushar Krishna
arXiv：https://arxiv.org/abs/2607.22465
发布时间：2026-07-24
更新时间：2026-07-27
arXiv 分类：cs.AI, cs.LG, cs.MA
关联标签：model-routing, contextual-bandit, cost-quality, agentic-workflows, deployment

English Abstract

Routing to select large language models (LLMs) with different cost-quality trade-offs has become a fundamental deployment feature of enterprise AI. Existing routers, primarily make independent routing decisions for each LLM call. However, agentic applications execute as long-horizon workflows whose quality is determined only by a delayed, task-level outcome. This mismatch prevents per-call routers from correctly attributing feedback to individual routing decisions. Towards mitigating this, we present TRACE-Router, a task-level routing framework that aligns routing with the unit of supervision. TRACE-Router assigns each task to a model once at admission using a contextual bandit, pins all subsequent LLM calls to the selected backend, and updates its policy using the task's terminal reward, jointly accounting for accuracy and latency. By leveraging delayed task feedback, TRACE-Router learns routing policies that adapt to the workload while avoiding explicit task-complexity estimation. Across three agentic benchmarks, TRACE-Router consistently improves the accuracy-latency trade-off, achieving non-dominated Pareto frontier points. On tau2-Bench, it outperforms latency-matched interpolation between individual models by 7-8 accuracy points, while on Terminal-Bench it achieves 7.1 higher accuracy points than the strongest single model baseline with 36% lower latency.

English Summary

TRACE-Router addresses the mismatch between per-call LLM routing and task-level outcomes in agentic workflows. Instead of routing each LLM call independently, it assigns a model once at task admission via a contextual bandit, pins all subsequent calls to that backend, and updates policy using the terminal reward (jointly accounting for accuracy and latency). On tau2-Bench it outperforms latency-matched interpolation by 7-8 accuracy points; on Terminal-Bench it achieves 7.1 higher accuracy than the strongest single model with 36% lower latency.

Obsidian Notes

内容由 opencli arxiv paper 拉取 arXiv 元数据与摘要生成。
中文导读与价值判断均锚定在条目已有摘要、论文摘要、作者、日期与分类信息上；未补充论文摘要之外的实验细节。