Back to Arena

MetaGPT

by DeepWisdom / geekan

System Card

OrganizationDeepWisdom / geekan
Released2023-01
Architectureagentic-workflow / SOP-driven multi-role memory
DetailsRoles (PM, architect, engineer, etc.) share an environment-level memory that stores messages, artifacts, and subscriptions. Code = SOP(Team) philosophy; iclr2024.
Parameters—
Domainagent-memory
Open SourceYes
PaperView Paper
WebsiteVisit
CodeRepository
sopmulti-agentsoftware-companyiclr-poster

Capability Profile

Benchmark Scores

5 of 14 benchmarks
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
1/2
LoCoMo
77.974p
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
77.761p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
70.236p
HotpotQA
74.767p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:MetaGPT paper (arXiv:2308.00352); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)MetaGPT paper (arXiv:2308.00352); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)MetaGPT paper (arXiv:2308.00352); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)MetaGPT paper (arXiv:2308.00352); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)MetaGPT paper (arXiv:2308.00352); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)