List your company
ResearchFormerly WizWand

Knowledge Base Question Answering on GrailQA Freebase (test)

86.9Hits@1SoG
4.563225.939147.31568.6909Sep 30, 2025Oct 28, 2025Nov 26, 2025Dec 25, 2025Jan 23, 2026Feb 21, 2026Mar 22, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SoGBackbone=GPT-4o2025.1086.9
Plan-on-GraphBackbone=GPT-42025.1084.7
ReKnoSBackbone=GPT-42025.1082.7
Think-on-GraphBackbone=GPT-42025.1081.4
Debate-on-GraphBackbone=GPT-42025.1080
EffiQABackbone=GPT-42025.1078.4
PoGBackbone Model=Qwen 3-235B2025.0977.93
ReknoSBackbone Model=Qwen 3-235B2025.0975.67
PoGBackbone Model=Qwen 2.5-72B2025.0975.3
ReknoSBackbone Model=Qwen 2.5-72B2025.0973.67
iQUESTBackbone=GPT-4o2025.1073.5
ToGBackbone Model=Qwen 2.5-72B2025.0971.53
PoGBackbone Model=Qwen 2.5-7B2025.0969.88
ReknoSBackbone Model=Qwen 2.5-7B2025.0969.58
KG-R1Backbone Model=Qwen2.5 3B, Training Dataset=WebQSP2025.0968.55
Think-on-GraphBackbone=GPT-4o, Reproduced status=reproduced2025.1065.5
KG-R1Backbone Model=Qwen2.5 3B, Training Dataset=CWQ2025.0964.84
GPT-4o-mini + KGStrategy=KG-Augmented without Fine-Tuning2026.0360
DeepSeek-R1-Distill-Llama-70B + KGSize=70B, Strategy=KG-Augmented without Fine-Tuning2026.0359.02
LLaMA-3.3-70B + KGSize=70B, Strategy=KG-Augmented without Fine-Tuning2026.0357.7
ToGBackbone Model=Qwen 2.5-7B2025.0954.73
ToGBackbone Model=Qwen 3-235B2025.0953.2
KG-Hopper w/Qwen-2.5-7BSize=7B, Strategy=KG-Augmented with RL Fine-Tuning2026.0350.1
LLaMA-3.1-8B (SFT) + KGSize=8B, Strategy=KG-Augmented with Supervised Fine-Tuning2026.0347.7
Qwen-2.5-7B (SFT) + KGSize=7B, Strategy=KG-Augmented with Supervised Fine-Tuning2026.0346.18
Qwen-2.5-7B + KGSize=7B, Strategy=KG-Augmented without Fine-Tuning2026.0341.1
LLaMA-3.1-8B + KGSize=8B, Strategy=KG-Augmented without Fine-Tuning2026.0340.4
Chain-of-ThoughtBackbone=GPT-4o2025.1038
Self-ConsistencyBackbone=GPT-4o2025.1037.4
GPT-4o-miniStrategy=LLM Prompting Only2026.0336.22
IO PromptBackbone=GPT-4o2025.1035.3
GPT-4oStrategy=LLM Prompting Only2026.0335.01
LLaMA-3.3-70BSize=70B, Strategy=LLM Prompting Only2026.0333.79
DeepSeek-R1-Distill-Llama-70BSize=70B, Strategy=LLM Prompting Only2026.0331.7
Qwen-2.5-7BSize=7B, Strategy=LLM Prompting Only2026.0329.53
LLaMA-3.1-8BSize=8B, Strategy=LLM Prompting Only2026.0328.35
RoGBackbone Model=Qwen 2.5-72B2025.0922.63
RoGBackbone Model=Qwen 3-235B2025.0918.1
RoGBackbone Model=Qwen 2.5-7B2025.0917.67
SCBackbone Model=Qwen 3-235B2025.0915.73
CoTBackbone Model=Qwen 3-235B2025.0914.6
IOBackbone Model=Qwen 3-235B2025.0912.8
IOBackbone Model=Qwen 2.5-72B2025.0911.63
SCBackbone Model=Qwen 2.5-72B2025.0911.43
CoTBackbone Model=Qwen 2.5-72B2025.0910.53
SCBackbone Model=Qwen 2.5-7B2025.098.6
IOBackbone Model=Qwen 2.5-7B2025.098.07
CoTBackbone Model=Qwen 2.5-7B2025.097.73