Multi agent relay thinking improves collaboration while saving resources
Multi-Agent Relay Thinking
Computer Vision and Pattern Recognition
Summary
Most multi-agent systems work by having each agent think separately and then combining their answers. The authors explore a different approach where agents pass their thinking to the next one, who continues and improves the reasoning step by step. They created a method called Multi-Agent Confidence-guided Relay (MACR) that decides when to pass the reasoning and which agent should take over based on confidence scores. Their tests show this approach can be more accurate than relying on single agents and as good as multi-agent methods while using fewer computational resources. The method also lets different types of agents work together, sharing their unique strengths over time.
What this means in practice
- •For ai system developers: Build multi-agent reasoning systems that dynamically hand off tasks to improve accuracy while reducing computation costs.
- •For chatbot platform teams: Implement relay-style reasoning among diverse AI agents to integrate complementary skills and enhance conversational problem solving.
Authors
Xu Liu, Jungang Qiao, Guikun Chen, Xiaojuan Qi, Wenguan Wang
Abstract
Existing multi-agent collaboration (MAC) methods have primarily explored a group thinking collaboration topology, where agents independently generate solutions and aggregate their outputs. In this work, we investigate an alternative relay thinking topology, where reasoning ownership is dynamically transferred across agents, allowing subsequent agents to inherit and refine an evolving reasoning trajectory. We introduce Multi-Agent Confidence-guided Relay (MACR), a framework that enables dynamical relay collaboration among agents. The key challenge is determining when reasoning should be relayed and which agent should continue the process. method addresses this challenge through confidence-guided relay triggering and a bidding mechanism that selects the next agent based on its confidence. Experiments on four benchmarks show that method outperforms the strongest single-model strategies in accuracy, e.g., +3.3% on MathVista and +3.9% on MMStar, and remains on par with multi-agent strategies while consuming 6.0x less token overhead on HallusionBench and MMStar.Further analysis demonstrates that relay collaboration enables heterogeneous agents to progressively integrate complementary insights, combining heterogeneous capabilities across reasoning stages. Our source code is available at https://github.com/67L1/MACR.