Back to Arena

KAG

by OpenSPG / Ant Group

System Card

OrganizationOpenSPG / Ant Group
Released2025-06
Architectureknowledge-base / Logical-form guided KG reasoning
DetailsThree-tier: kg-builder (DIKW hierarchy extraction with schema constraint), kg-solver (hybrid planning + retrieval + reasoning operators), kag-model. Supports text retrieval, KG reasoning, and numerical calc in one pipeline.
Parameters—
Domainknowledge-graphrag-retrieval
Open SourceYes
PaperView Paper
WebsiteVisit
CodeRepository
logical-formschema-constrainedant-groupprofessional-domain

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
2/5
RULER
72.563p
NIAH
no data
LooGLE
no data
LongBench
603p
∞Bench
no data
Multi-Turn Recall
0/2
LoCoMo
no data
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
71.529p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
74.373p
HotpotQA
76.278p
Agent Task Memory
0/1
AgentBench-Mem
no data
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
1/1
RAGAS
74.393p
Sources:arXiv:2409.13731 Table 8 — F1 with LFSH_ref3 + DeepSeek-V2 backboneKAG paper (arXiv:2409.13731); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)KAG paper (arXiv:2409.13731); evaluated on RAGAS: Automated Evaluation of Retrieval-Augmented Generation (Exploding Gradients, 2309)KAG paper (arXiv:2409.13731); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)KAG paper (arXiv:2409.13731); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)KAG paper (arXiv:2409.13731); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)