MASBench: Benchmarking LLM-based Multi-Agent Collaboration under Partial Observability

DOI10.48550/arXiv.2610.04672MaRDI QIDQ7367870arXivFDO

Authors Zekai Yu, Sijie Wen, Qizhi Chu, Yang Liu, Chen Qian, Cheng Yang, Chuan Shi, Zhiyuan Liu

Publication date 3 October 2026

Published in arXiv

Tldr We introduce MASBench, a multi-agent collaboration benchmark designed under partially observable constraints to systematically evaluate collaboration mechanisms in Large Language Model (LLM)-based Multi-Agent Systems (MAS). It consists of three task categories and provides deterministic evaluation metrics for collaboration outcomes and communication overhead.

Full work available at URL https://doi.org/10.48550/arXiv.2610.04672

arXiv classification

cs.AI

Cites work

• A concise introduction to decentralized POMDPs
• (DATASET) cloud
• (SOFTWARE) openai
• The Complexity of Decentralized Control of Markov Decision Processes

This page was built for publication: MASBench: Benchmarking LLM-based Multi-Agent Collaboration under Partial Observability

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7367870)