智能调度大模型，省钱提效还更准
MTRouter: Cost-Aware Multi-Turn LLM Routing with History-Model Joint Embeddings
- 用历史与模型联合嵌入，动态决定每轮调哪个模型
- 在科学和考试任务上成本降43%~59%，效果超GPT-5
- 减少切换次数，抗错强，还能让模型各司其职
多轮长周期任务在大语言模型中日益普遍，但需多次调用模型，导致推理成本高昂。本文研究成本约束下的多轮模型调度问题：在固定预算下，从模型池中选择每轮调用的模型。提出MTRouter，将交互历史与候选模型编码为联合嵌入，并通过日志轨迹学习结果预测器，估算每轮模型的效用。实验表明，MTRouter显著优化性能与成本的权衡：在ScienceWorld上，性能超越GPT-5的同时总成本降低58.7%；在Humanity's Last Exam（HLE）上，达到可比准确率，总成本较GPT-5降低43.4%，且在未见任务上仍具优势。进一步分析揭示其有效性机制：相比以往方法，MTRouter减少模型切换，更耐受瞬时错误，并表现出模型间的涌现专业化特征。
原文摘要 · Abstract (English)
Multi-turn, long-horizon tasks are increasingly common for large language models (LLMs), but solving them typically requires many sequential model invocations, accumulating substantial inference costs. Here, we study cost-aware multi-turn LLM routing: selecting which model to invoke at each turn from a model pool, given a fixed cost budget. We propose MTRouter, which encodes the interaction history and candidate models into joint history-model embeddings, and learns an outcome estimator from logged trajectories to predict turn-level model utility. Experiments show that MTRouter improves the performance-cost trade-off: on ScienceWorld, it surpasses GPT-5 while reducing total cost by 58.7%; on Humanity's Last Exam (HLE), it achieves competitive accuracy while reducing total cost by 43.4% relative to GPT-5, and these gains even carry over to held-out tasks. Further analyses reveal several mechanisms underlying its effectiveness: relative to prior multi-turn routers, MTRouter makes fewer model switches, is more tolerant to transient errors, and exhibits emergent specialization across models. Code: https://github.com/ZhangYiqun018/MTRouter
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。