More Capable, Less Cooperative? When LLMs Fail At Zero-Cost Collaboration

2026-04-09

AI Summary

This study investigates why large language models fail to cooperate even in zero-cost collaborative settings—where assisting others incurs no loss or gain and explicit cooperation is requested. By constructing a simplified multi-agent environment and integrating causal decomposition, communication interventions, and reasoning trace analysis, the work reveals for the first time that model capability exhibits no positive correlation with willingness to cooperate. The authors demonstrate that explicit coordination protocols and minimal shared incentives substantially enhance collaborative performance: under such protocols, a lower-capability model (o3-mini) achieves 50% of optimal collective performance, whereas a stronger model (o3) reaches only 17%. Moreover, even slight incentives effectively mitigate weak cooperative tendencies, underscoring the necessity of purpose-built mechanisms to foster reliable collaboration in artificial agents.

Abstract

Large language model (LLM) agents increasingly coordinate in multi-agent systems, yet we lack an understanding of where and why cooperation failures may arise. In many real-world coordination problems, from knowledge sharing in organizations to code documentation, helping others carries negligible personal cost while generating substantial collective benefits. However, whether LLM agents cooperate when helping neither benefits nor harms the helper, while being given explicit instructions to do so, remains unknown. We build a multi-agent setup designed to study cooperative behavior in a frictionless environment, removing all strategic complexity from cooperation. We find that capability does not predict cooperation: OpenAI o3 achieves only 17% of optimal collective performance while OpenAI o3-mini reaches 50%, despite identical instructions to maximize group revenue. Through a causal decomposition that automates one side of agent communication, we separate cooperation failures from competence failures, tracing their origins through agent reasoning analysis. Testing targeted interventions, we find that explicit protocols double performance for low-competence models, and tiny sharing incentives improve models with weak cooperation. Our findings suggest that scaling intelligence alone will not solve coordination problems in multi-agent systems and will require deliberate cooperative design, even when helping others costs nothing.

Problem

• large language models
• multi-agent systems
• cooperation failure
• zero-cost collaboration
• collective performance

Innovation

• zero-cost collaboration
• multi-agent cooperation
• causal decomposition
• cooperative design
• LLM coordination failure