List your company
ResearchFormerly WizWand

Question Answering on NQ (Exact Match)

74.2Exact MatchFreshPER
2.02420.76239.558.238Oct 17, 2025Nov 22, 2025Dec 28, 2025Feb 2, 2026Mar 10, 2026Apr 15, 2026May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
FreshPERType=LLM2026.0474.2
Neuro-RITRanking Strategy=Reranker2026.0472.57
PA-RAGRanking Strategy=Reranker2026.0472.01
Neuro-RIT + RerankerGenerator=Mistral-7B-Instruct-v0.22026.0469.79
Neuro-RITRanking Strategy=None2026.0469.68
RALM + RerankerGenerator=Mistral-7B-Instruct-v0.22026.0468.48
PA-RAGRanking Strategy=None2026.0467.92
Neuro-RITGenerator=Mistral-7B-Instruct-v0.22026.0466.47
PA-RAGRanking Strategy=RankCoT2026.0466.3
RALMGenerator=Mistral-7B-Instruct-v0.22026.0464.85
InstructRAGRanking Strategy=Reranker2026.0463.58
Neuro-RIT + RankCoTGenerator=Mistral-7B-Instruct-v0.22026.0463.48
RALMRanking Strategy=None2026.0462.21
Neuro-RITRanking Strategy=RankCoT2026.0461.93
RALMRanking Strategy=Reranker2026.0461.65
InstructRAGRanking Strategy=None2026.0460.62
RALM + RankCoTGenerator=Mistral-7B-Instruct-v0.22026.0460.31
InstructRAGRanking Strategy=RankCoT2026.0460.17
RALMRanking Strategy=RankCoT2026.0459.43
RetRobustRanking Strategy=Reranker2026.0454.42
RetRobustRanking Strategy=RankCoT2026.0453.26
RetRobustRanking Strategy=None2026.0453.23
On-PolicyType=LLM2026.0450.8
SLATEBackbone=Qwen2.5-7B-Base, training_data=NQ+HotpotQA2026.0249.7
SEARCH-R1Backbone=Qwen2.5-7B-Base, training_data=NQ+HotpotQA2026.0248
Search-E1-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Process-supervision RL, Base Model=Qwen2.5-3B2026.0547.4
AutoRefine-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL, Base Model=Qwen2.5-3B2026.0546.7
EvolveRBackbone Model=Qwen2.5-7B2025.1046.2
PiCABackbone=Qwen2.5-7B Instruct2026.0546
ZeroSearchBackbone=Qwen2.5-7B Instruct2026.0543.6
AutoRefine-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL, Base Model=Qwen2.5-3B2026.0543.6
TIPSBackbone=Qwen2.5-3B Instruct2026.0543.5
EvolveRBackbone Model=Qwen2.5-3B2025.1043.4
TIPSBackbone=Qwen2.5-7B Instruct2026.0543.4
ReSearch-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL, Base Model=Qwen2.5-3B2026.0542.7
PiCABackbone=Qwen2.5-3B Instruct2026.0542.6
SLATEBackbone=Qwen2.5-3B-Base, Training Data=NQ+HotpotQA2026.0242.5
MT-PPOBackbone=Qwen2.5-7B Instruct2026.0542.4
Search-R1-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL, Base Model=Qwen2.5-3B2026.0542.1
GiGPO-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Process-supervision RL, Base Model=Qwen2.5-3B2026.0542
ZeroSearchBackbone=Qwen2.5-3B Instruct2026.0541.4
ZeroSearchBackbone=Qwen2.5-7B-Base2026.0241.1
SEARCH-R1Backbone=Qwen2.5-3B-Base, Training Data=NQ+HotpotQA2026.0240.6
Search-R1-baseBackbone Model=Qwen2.5-3B2025.1040.6
ReSearchBackbone=Qwen2.5-7B-Base2026.0239.8
MT-PPOBackbone=Qwen2.5-3B Instruct2026.0539.7
Search-R1-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL, Base Model=Qwen2.5-3B2026.0539.7
Search-R1-instructBackbone Model=Qwen2.5-7B2025.1039.3
Search-R1Backbone=Qwen2.5-7B Instruct2026.0539.3
RAGBackbone=Qwen2.5-7B-Base2026.0237.1
ReSearch-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL, Base Model=Qwen2.5-3B2026.0536.5
Rejection SamplingBackbone Model=Qwen2.5-7B2025.1036
SFTBackbone=Qwen2.5-7B-Base2026.0235.3
RAGBackbone Model=Qwen2.5-7B2025.1034.9
RAGBackbone=Qwen2.5-7B Instruct2026.0534.9
ZeroSearchBackbone=Qwen2.5-3B-Base2026.0234.8
RAGBackbone Model=Qwen2.5-3B2025.1034.8
RAGBackbone=Qwen2.5-3B Instruct2026.0534.8
Naive RAGRetrieval Strategy=Single-Hop Retrieval, Base Model=Qwen2.5-3B2026.0534.8
IRCoTBackbone=Qwen2.5-7B-Base2026.0234.2
Search-R1-instructBackbone Model=Qwen2.5-3B2025.1034.1
Search-R1Backbone=Qwen2.5-3B Instruct2026.0534.1
ReSearchBackbone=Qwen2.5-3B-Base2026.0233.9
Standard PERType=LLM2026.0433.6
Search-o1Backbone=Qwen2.5-7B-Base2026.0232.1
SFTBackbone Model=Qwen2.5-7B2025.1031.8
RAGBackbone=Qwen2.5-3B-Base2026.0231.2
SFTBackbone=Qwen2.5-3B-Base2026.0230.1
R1 (no search)Backbone=Qwen2.5-7B-Base, search=disabled2026.0229.8
R1-baseBackbone Model=Qwen2.5-7B2025.1029.7
Rejection SamplingBackbone Model=Qwen2.5-3B2025.1029.4
Qwen2.5-VL 3BText Source=Pure-Text, CEPE protocol=true2025.1029.31
IRCoTBackbone=Qwen2.5-3B-Base2026.0228.6
R1-instructBackbone Model=Qwen2.5-7B2025.1027
Search-o1Backbone=Qwen2.5-3B-Base2026.0226.7
R1 (no search)Backbone=Qwen2.5-3B-Base2026.0225.8
SFTBackbone Model=Qwen2.5-3B2025.1024.9
SFTRetrieval Strategy=None, Base Model=Qwen2.5-3B2026.0524.9
SEETOKText Source=Visual-Text, CEPE protocol=true2025.1024.14
Search-o1Backbone Model=Qwen2.5-3B2025.1023.8
Search-o1Backbone=Qwen2.5-3B Instruct2026.0523.8
Search-o1Retrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Prompting, Base Model=Qwen2.5-3B2026.0523.8
CoTBackbone=Qwen2.5-7B-Base2026.0222.8
R1-baseBackbone Model=Qwen2.5-3B2025.1022.6
R1-BaseRetrieval Strategy=None, Base Model=Qwen2.5-3B2026.0522.6
IRCoTBackbone Model=Qwen2.5-7B2025.1022.4
DirectBackbone=Qwen2.5-7B-Base2026.0221.3
Qwen2.5-VL 3BText Source=Visual-Text, CEPE protocol=true2025.1021.13
R1-instructBackbone Model=Qwen2.5-3B2025.1021
R1-InstructRetrieval Strategy=None, Base Model=Qwen2.5-3B2026.0521
CoTBackbone=Qwen2.5-3B-Base2026.0216.8
DirectBackbone=Qwen2.5-3B-Base2026.0215.2
Search-o1Backbone Model=Qwen2.5-7B2025.1015.1
Search-o1Backbone=Qwen2.5-7B Instruct2026.0515.1
Direct InferenceBackbone Model=Qwen2.5-7B2025.1013.4
IRCoTBackbone Model=Qwen2.5-3B2025.1011.1
IRCoTRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Prompting, Base Model=Qwen2.5-3B2026.0511.1
Direct InferenceBackbone Model=Qwen2.5-3B2025.1010.6
Direct GenerationRetrieval Strategy=None, Base Model=Qwen2.5-3B2026.0510.6
CoTBackbone Model=Qwen2.5-7B2025.104.8