Back to Arena

AgentScope

by ModelScope (Alibaba)

System Card

OrganizationModelScope (Alibaba)
Released2024-01
Architectureagentic-workflow / Production agent framework with pluggable memory
DetailsProduction-ready agent framework with InMemoryMemory, database-backed memory, memory compression, long-term memory via ReMe integration, SQLite-session persistence, OpenTelemetry. ReAct agents and MCP-native.
Parameters—
Domainagent-memorylong-context
Open SourceYes
PaperView Paper
WebsiteVisit
CodeRepository
alibabaproductionreme-integrationmemory-compression

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
2/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
603p
∞Bench
70.39p
Multi-Turn Recall
1/2
LoCoMo
7547p
MemoryBank
no data
Cross-Session Memory
0/1
LongMemEval
no data
Multi-Hop QA
2/3
BABILong
75.652p
MultiHop-RAG
no data
HotpotQA
75.371p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:AgentScope paper (arXiv:2402.14034); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)AgentScope paper (arXiv:2402.14034); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)AgentScope paper (arXiv:2402.14034); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)AgentScope paper (arXiv:2402.14034); evaluated on InfiniteBench: Extending Long Context Evaluation Beyond 100K Tokens (Tsinghua / OpenBMB, 2402)AgentScope paper (arXiv:2402.14034); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)AgentScope paper (arXiv:2402.14034); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)