MANTA Lets Multi-Agent Communication Topology Rewire Itself Mid-Task, Scoring 74.0 Average Across Five Benchmarks

arXiv·medium signalXBlueskyLinkedInCopy link

Most LLM multi-agent systems treat communication topology as a fixed design choice or an offline optimization target. MANTA initializes a task-conditioned topology from prior structural experience, then monitors collaboration traces during deployment and applies bounded structural updates — modifying agent roles, communication links, execution order, information visibility, and validation pathways — while preserving the task interface and agent budget. Across five benchmarks spanning information seeking, tool use, planning, workflow execution, and mathematical reasoning it reaches a 74.0 average, 5.8 percentage points above the strongest baseline, with the best result on PlanCraft.