智能体
x.com ·
@elvis(@omarsar0)
AgentWorld 论文：多智能体团队有效动作不足三分之一，协调任务成功率仅 12%
AgentWorld 将 3 至 20 个不同角色的 LLM agent 放入游戏沙箱执行 50 轮以上任务，agent 之间无法看到彼此内部状态，只能通过消息和共享计划协调。结果显示多智能体团队中真正有助于完成任务的行动不足三分之一，Gemini 3 Flash 任务成功率最高为 52.0%，协调类任务最难，成功率仅 12%，常见失败包括通信中断、角色混淆和共享计划丢失。论文编号 arXiv 2609.31590。
AI推荐理由 引用 AgentWorld 论文（arXiv 2609.31590）给出具体数据：3-20 个 LLM agent 在 50+ 轮任务中，多智能体团队实际有效动作不足三分之一，Gemini 3 Flash 任务成功率 52.0%，协调类任务仅 12%，失败模式包括通信中断、角色混淆和共享计划丢失。这是对多智能体扩展性的可验证反直觉证据，对 Agent 架构设计有直接参考价值。局限：数据来自单一 benchmark，推文未提供论文细节，需以原文为准。
78
正文
More agents don't mean higher performance.
There is a coordination bottleneck to consider. Not to mention the unnecessary costs.
So how many of a multi-agent team's actions actually help it finish the task?
In this AgentWorld paper, fewer than a third.
AgentWorld puts 3 to 20 LLM agents with different roles into a game sandbox for tasks that run 50+ rounds.
Agents can't see each other's internal state, so they have to coordinate through messages and shared plans.
Gemini 3 Flash has the highest task success at 52.0%. Coordination tasks are the hardest category, at 12% success, and common failures include communication breakdowns, role confusion, and lost shared plans.
Paper: [arxiv.org/abs/2609.31590](https://arxiv.org/abs/2609.31590)
Chat with Paper: [academy.dair.ai/papers/agentwo…](https://academy.dair.ai/papers/agentworld-benchmarking-long-horizon-collaboration-of-multi-agent-llms-2609.31590)
💬0🔄0❤️0