研究库 论文知识库

内容库 / 论文

arXiv:2609.31590 · Agent 智能体

AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs

AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs

类型：arxiv
标识：2609.31590
链接：https://arxiv.org/abs/2609.31590
主分类：agent
形态：benchmark
TLDR：Existing multi-agent benchmarks primarily test in competitive settings, short-horizon interactions under 20 steps, or simply aggregate individual performance, failing to isolate and highlight genuine collaboration capabilities of LLM-based agents. We introduce AgentWorld, a benchmark of 100 human-annotated tasks (with 100 augmented variants) for evaluating long-horizon, multi-agent collaboration. Tasks span 50+ interaction rounds across a rich MMORPG sandbox and require 3-20 agents with asymmetric roles and abilities to coordinate through communication, joint planning, and resource sharing und
副分类：evaluation
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-09-28-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

笔记 Tom 文献雷达 · Agent/RAG/长上下文 · 2026-08-03 08:40 UTC

笔记 📋 Tom 文献雷达 · Agent + RAG + 长上下文 · 2026-09-01

笔记 Tom 文献雷达 · AI Agent / RAG / 长上下文 · 2026-09-05 14:40

笔记 Tom 文献雷达 · Agent RAG LongContext · 2026-09-04 08:40

笔记 Tom 文献雷达 · Agent + RAG + LongContext · 2026-08-29

笔记深度精读 · SoK: Agentic Retrieval-Augmented Generation（arXiv:2603.07379）

arXiv

2609.31590

TLDR

Existing multi-agent benchmarks primarily test in competitive settings, short-horizon interactions under 20 steps, or simply aggregate individual performance, failing to isolate and highlight genuine collaboration capabilities of LLM-based agents. We introduce AgentWorld, a benchmark of 100 human-annotated tasks (with 100 augmented variants) for evaluating long-horizon, multi-agent collaboration. Tasks span 50+ interaction rounds across a rich MMORPG sandbox and require 3-20 agents with asymmetric roles and abilities to coordinate through communication, joint planning, and resource sharing und

arXiv ↗ 中文全文