LLM-based multi-agent systems (MAS) have emerged as a promising approach to tackle complex tasks that are difficult for individual LLMs. A natural strategy is to scale performance by increasing the number of agents; however, we find that such scaling exhibits strong diminishing returns in homogeneous settings, while introducing heterogeneity (e.g., different models, prompts, or tools) continues to yield substantial gains. This raises a fundamental question: what limits scaling, and why does diversity help? We present an information-theoretic framework showing that MAS performance is bounded by the intrinsic task uncertainty, not by agent count. We derive architecture-agnostic bounds demonstrating that improvements depend on how many effective channels the system accesses. Homogeneous agents saturate early because their outputs are strongly correlated, whereas heterogeneous agents contribute complementary evidence. We further introduce $K^*$, an effective channel count that quantifies the number of effective channels without ground-truth labels. Empirically, we show that heterogeneous configurations consistently outperform homogeneous scaling: 2 diverse agents can match or exceed the performance of 16 homogeneous agents. Our results provide principled guidelines for building efficient and robust MAS through diversity-aware design. Code and Dataset are available at the link: https://github.com/SafeRL-Lab/Agent-Scaling.
AgentInit: Initializing LLM-based Multi-Agent Systems via Diversity and Expertise Orchestration for Effective and Efficient Collaboration
Institute of Computing and Intelligence, Harbin Institute of Technology, Shenzhen(哈尔滨工业大学(深圳)计算与智能研究院);The University of Sydney(悉尼大学);School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen(哈尔滨工业大学(深圳)计算机科学与技术学院)
Constructive Conflict-Driven Multi-Agent Reinforcement Learning for Strategic Diversity
School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院);CRISE, Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所);The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences(复杂系统认知与决策智能重点实验室,中国科学院自动化研究所)
Beyond Symmetric Agents: Cognitive Diversity and Multi-Agent Debate in Small Language Models
Boston Children’s Hospital, Harvard Medical School(波士顿儿童医院,哈佛医学院);Harvard School of Engineering and Applied Sciences, Harvard College(哈佛工程与应用科学学院,哈佛学院)
Dynamic Knowledge Exchange and Dual-diversity Review: Concisely Unleashing the Potential of a Multi-Agent Research Team
East China Normal University(华东师范大学);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Peking University(北京大学);University of Chinese Academy of Sciences(中国科学院大学);Bytedance Inc(字节跳动);Canadian Collaborative Entrepreneurs Organization Inc(加拿大合作企业家组织)
EduAgentQG: Multi-Agent Personalized Mathematics Question Generation with Explicit Diversity and Objective-Aware Evaluation
East China Normal University(东华大学);Shanghai Institute of Al for Education(上海人工智能教育研究院);The Chinese University of Hong Kong, Shenzhen School of Data Science(香港中文大学(深圳)数据科学学院)
Phasic Diversity Optimization for Population-Based Reinforcement Learning
Dalian University of Technology(大连理工大学);Northwestern Polytechnical University(西北工业大学);Yangzhou Collaborative Innovation Research Institute Co., Ltd(扬州协同创新研究院有限公司);Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所)
Diversity of Transformer Layers: One Aspect of Parameter Scaling Laws
Nara Institute of Science and Technology(奈良科学技術大學);RIKEN Center for Advanced Intelligence Project(RIKEN先進 Intelligenceプロジェクト);Chungnam National University(Chungnam國立大學);The University of Tokyo(東京大學);Institute of Science Tokyo(東京科學研究院)