Skip to main content
Search

# Multi-Agent Collaboration

4 related articles

Subscribe to updates
  1. arXiv

    Answering belief questions is not collaborating: Claude recovers every agent belief but returns optimal plans in only 67.9% of matched cases

    Theory-of-mind benchmarks usually ask story questions about what others believe, and a correct answer does not show the model can use that judgment in a task. This evaluation extends household collaboration and calendar scheduling with asymmetric information and separates world understanding, information access, beliefs, decisions and execution into independent interfaces. On matched scheduling cases Claude recovers all agent beliefs yet produces subjectively optimal plans in only 67.9% of cases, while supplying beliefs raises Qwen-27B household task success from 55.0% to 75.0%.

    Shenzhen Research Institute of Big DataByteDance (China)
    Zhenshuai Yin（First author）
    Theory-of-Mind ReasoningC-WAHCalBenchMulti-Agent CollaborationQwen3.5
  2. arXiv

    Multi-agent memory stops storing steps alone: task dependency graphs become the shared script, topping all six benchmark settings

    Multi-agent memory is reorganized around task dependency graphs: a shared blackboard records evidence during execution and corrects goal, plan, and world-state divergence, then verified steps and dependencies are distilled into reusable procedures, giving the highest task score in all six benchmark settings across ALFWorld, WebArena, and OfficeBench.

    China Telecom (China)Beijing University of Posts and TelecommunicationsChina Telecom
    Yi Liu（First author）
    Self-Evolving Procedural MemoryTask Dependency GraphDivergence-Aware Multi-Agent AlignmentMulti-Agent CollaborationAgent Memory
  3. arXiv

    Global Coherence in Multi-Agent Collaboration: Local Intelligence Cannot Substitute for Missing Global State

    The work defines the global coherence problem, in which AI agents each make locally valid decisions yet jointly produce an invalid result, and gives the Observation-Aliasing Impossibility Theorem: when k indistinguishable worlds require pairwise-disjoint actions, the best randomized worst-case success is 1/k; it then proposes local-to-global runtime semantics X = (H, C, G, F; D), where the harness owns shared state and governs commit, and nine studies show that when the deciding event is hidden a frontier model scores 12-17/40 (consistent with chance 1/3), restoring one authoritative fact returns 40/40, commit enforcement on TeamBench leaves 0/5 violations versus 5/5, and on tau2-bench Telecom the harness scores 1.00 while current-state checks score 0.07.

    Xin Heng（First author, Corresponding author）
    Global Coherence ProblemLocal-to-Global Runtime SemanticsMulti-Agent Collaboration
  4. arXiv

    AgentWorld tests multi-agent collaboration on 100 long-horizon MMORPG tasks: the best model reaches only 52.0% success and a CCE of 0.320

    The authors built AgentWorld, a benchmark of 100 human-annotated tasks (plus 100 LLM-augmented variants) on the open-source MMORPG engine Kaetram, requiring 3–20 agents with asymmetric roles to coordinate over 25–55 rounds under a blackbox setting, and proposed Causal Collaboration Effectiveness (CCE), a graph-based metric over causal action graphs; testing Gemini 3 Flash, Claude Haiku 4.5, GPT-5 Mini, and DeepSeek R1-70B, the best model reaches only 52.0% task success and a CCE of 0.320, with failure modes centered on communication breakdowns, role confusion, and inability to maintain shared plans across rounds.

    University of PennsylvaniaColumbia UniversitySeoul National UniversityPennsylvania State University
    Raphael Shu（First author）
    AgentWorldCausal Collaboration EffectivenessMulti-Agent CollaborationLLM Agents

You’re all caught up