RADAR: Redundancy-Aware Diffusion for Multi-Agent Communication Structure Generation

Compared with individual agents, large language model based multi-agent systems have shown great capabilities consistently across diverse tasks, including code generation, mathematical reasoning, and planning, etc. Despite their impressive performance, the effectiveness and robustness of these systems heavily rely on their communication topology, which is often fixed or generated in a single step. This restricts fine-grained structural exploration and flexible composition, resulting in excessive token utilization on simple tasks while limiting capability on complicated tasks. To mitigate this challenge, we introduce RADAR, a redundancy-aware and query-adaptive generative framework that actively reduce communication overhead. Motivated by recent progress in conditional discrete graph diffusion models, we formulate communication topology design as a step-by-step generation process, guided by the effective size of the graph. Comprehensive experiments on six benchmarks demonstrate that RADAR consistently outperforms recent baselines, achieving higher accuracy, lower token consumption, and greater robustness across diverse scenarios. Our code and data are available at https://github.com/cszhangzhen/RADAR.

Large language model based multi-agent systems — where several AI agents work together to solve a problem — can perform much better than a single AI model on tasks like coding, math, and planning. However, these systems often rely on fixed communication patterns, meaning the agents always exchange information in the same way regardless of how difficult the task is. This can waste large amounts of computation on simple problems while still struggling on more complex ones. To address this issue, we introduce RADAR, a framework that dynamically designs how AI agents communicate with each other for each individual query. Instead of creating the entire communication structure at once, RADAR builds it step by step, allowing the system to adaptively decide which agents should interact and how much collaboration is actually needed. The framework also reduces redundant communication to improve efficiency. We evaluated RADAR on six challenging benchmarks covering different reasoning and decision-making tasks. The results show that RADAR not only achieves higher accuracy than existing approaches, but also uses fewer tokens and remains more robust under diverse settings. This suggests that adaptive communication can make collaborative AI systems both smarter and more efficient.