Multi-agent systems reduce costly reasoning for faster collaboration
Token-Efficient Multi-Agent Collaboration via System One-Guided Computational Division of Labor
Multiagent SystemsArtificial Intelligence
Summary
Coordinating many AI agents to work together can get slow and expensive because the system uses powerful language models even for simple control tasks. The authors created a new method that uses a lightweight 'System One' model for managing coordination, while the stronger models focus on complex reasoning. This way, the system saves a lot of time and computational resources without losing accuracy. Their tests show this approach cuts costs and speeds up problem-solving in various tasks.
What this means in practice
- •For web service developers: Implement scalable multi-agent frameworks that significantly reduce API costs and response times when using large language models.
- •For automation engineers: Design multi-agent automation systems that dynamically divide work between lightweight controllers and powerful reasoning agents for efficient task completion.
Authors
Zihan Zhou, Xinzhe Hu, Hanxu Yang, Liangjian Wen, Zhao Kang
Abstract
Large language model (LLM)-based multi-agent systems (MAS) have become a promising paradigm for complex information-seeking and reasoning tasks by enabling collaborative problem solving among specialized agents. However, existing MAS frameworks tightly couple task reasoning with coordination operations, including task selection, role assignment, message routing, and context management. As interactions grow, using powerful LLMs for these bounded control decisions introduces substantial token overhead and latency, limiting the scalability of agentic Web services. In this paper, we investigate whether coordination can be decoupled from expensive reasoning without compromising collaborative performance. We propose S1-MAS, a token-efficient multi-agent framework based on System One-guided computational division of labor. S1-MAS assigns bounded coordination decisions to lightweight System One models while reserving open-ended reasoning for capable LLM workers. Specifically, a lightweight controller selects inspection conditions, chooses subsequent tasks, and determines termination, while a compact reader retrieves condition-relevant evidence from authorized sources to support these decisions. Through a decision-evidence loop, selected tasks dynamically determine worker roles and source access, enabling adaptive collaboration without task-specific training. Extensive experiments on seven diverse benchmarks demonstrate that S1-MAS achieves superior accuracy while substantially reducing the inference cost. Across individual comparisons with AgentVerse, DyLAN, and SelfOrg on seven benchmarks, S1-MAS reduces GPT-4o token consumption by 44.9%-97.2% and measured end-to-end latency by 37.8%-93.0%. These results highlight its potential for scalable and cost-effective agentic Web applications.