一篇最新的论文通过大量实验来验证什么时候多智能体系统（MAS）比单智能体（SAS）更好。
文章对比了 SAS 基线以及四种 MAS 架构：
- 中心式：由一个中心 agent 进行任务协调；
- 去中心式：agent 之间直接通信；
- 独立式：多个 agent 独立解决任务，最后统一结果；
- 混合式：同时有中心 agent 和 p2p 通信。
并在 6 个 benchmark 上进行实验：
- Finance Agent：金融分析；
- BrowseComp-Plus：网页搜索；
- WorkBench：现实工作场景任务；
- SWE-bench Verified：软件工程任务；
- Terminal-Bench：终端任务；
- PlanCraft：游戏规划。
主要结论如下：
1. MAS 没有普遍优势，在不同 benchmark 上的结果极其分化，在 Finance Agent 上可以达到 80.8% 的提升，而 PlanCraft 上会导致 -70%～-39% 的负收益。
2. SAS 基线的表现是判断是否应该使用 MAS 的最有用的标准，当 SAS 已经足够强时，继续加入 agent 很可能带来零收益甚至负收益，而实验得出的成功率阈值是 45%。
3. 任务的可拆解性和复杂性共同影响 MAS 效率。MAS 在可拆解的任务中可以获得很大收益，但在高复杂度的线性任务中往往会导致退化。
4. MAS 会导致错误信息的传播进而影响表现，不同 MAS 架构对于错误传播的阻断效果不同，中心式架构最能阻断错误传播，而独立式架构最难阻断错误传播。
这篇论文还有很多有意思的结论，我就不一一列举了，感兴趣的小伙伴可以去看原文，原文做了大量的实验和严谨的检验。
原文链接：https://www.nature.com/articles/s42256-026-01268-y
第一次做这种内容感觉很粗糙，后续我会再发布一些多智能体系统以及智能体自进化相关的内容。
Capable language models can outgrow the benefits of collaboration - Nature Machine IntelligenceAgents, language model-based systems that can reason, plan and act with tools to accomplish tasks, are widely deployed, yet it remains unclear when multi-agent coordination outperforms a strong single agent. Here we conduct a controlled experiment that holds task prompts, tools and compute budgets constant while varying only coordination structure and model capability. Across 260 configurations spanning six benchmarks, five architectures and three LLM families, we derive a predictive model using empirical coordination metrics. Across benchmarks, single-agent baseline performance emerges as the most robust predictor of whether coordination improves or decreases performance. In particular, we identify an empirical capability-saturation threshold beyond which additional agents are unlikely to improve performance. This threshold correctly predicts the effect of multi-agent coordination on performance in 94% of validation configurations on SWE-bench Verified and Terminal-Bench. We therefore interpret this threshold as a practical selection rule rather than a universal scaling principle. A second effect, baseline-scaled error amplification, survives cluster-robust inference (Probust = 0.030) and supports the failure-mode taxonomy. The fitted model achieves cross-validated R2 = 0.373 (0.413 with a task-grounded capability metric) and selects the best architecture in 87% of held-out configurations. These results provide a quantitative framework for within-domain architecture selection and for estimating when multi-agent coordination is likely to improve performance or add overhead. A controlled study of large language model agents across 260 configurations shows when multi-agent collaboration helps or hurts performance, and introduces a predictive model that selects the best architecture in 87% of held-out within-domain configurations.https://nature.com/articles
Capable language models can outgrow the benefits of collaboration - Nature Machine IntelligenceAgents, language model-based systems that can reason, plan and act with tools to accomplish tasks, are widely deployed, yet it remains unclear when multi-agent coordination outperforms a strong single agent. Here we conduct a controlled experiment that holds task prompts, tools and compute budgets constant while varying only coordination structure and model capability. Across 260 configurations spanning six benchmarks, five architectures and three LLM families, we derive a predictive model using empirical coordination metrics. Across benchmarks, single-agent baseline performance emerges as the most robust predictor of whether coordination improves or decreases performance. In particular, we identify an empirical capability-saturation threshold beyond which additional agents are unlikely to improve performance. This threshold correctly predicts the effect of multi-agent coordination on performance in 94% of validation configurations on SWE-bench Verified and Terminal-Bench. We therefore interpret this threshold as a practical selection rule rather than a universal scaling principle. A second effect, baseline-scaled error amplification, survives cluster-robust inference (Probust = 0.030) and supports the failure-mode taxonomy. The fitted model achieves cross-validated R2 = 0.373 (0.413 with a task-grounded capability metric) and selects the best architecture in 87% of held-out configurations. These results provide a quantitative framework for within-domain architecture selection and for estimating when multi-agent coordination is likely to improve performance or add overhead. A controlled study of large language model agents across 260 configurations shows when multi-agent collaboration helps or hurts performance, and introduces a predictive model that selects the best architecture in 87% of held-out within-domain configurations.https://nature.com/articles