List your company

Question Answering on NQ (F1 Score)

78.8F1 Score (NQ)Tri-RAG
12.18829.481546.77564.0685Aug 8, 2025Sep 22, 2025Nov 7, 2025Dec 23, 2025Feb 7, 2026Mar 25, 2026May 10, 2026

Evaluation results

MethodLinks
Tri-RAGGenerator Backbone=Qwen-Max2026.0478.8
Tri-RAGGenerator Backbone=Qwen-Long2026.0478.7
AflowGenerator Backbone=Qwen-Long2026.0474.6
AflowGenerator Backbone=Qwen-Max2026.0473.9
SafeChain2026.0473.9
Tri-RAGGenerator Backbone=Gemma-2-27B2026.0472.5
STAR-12026.0472.3
R1-ALT2026.0472
No train2026.0471.7
Tri-RAGGenerator Backbone=GPT-4o-mini2026.0469.8
Tri-RAGGenerator Backbone=GPT-4.1-nano2026.0469.8
AflowGenerator Backbone=Gemma-2-27B2026.0469.4
Self-ActGenerator Backbone=Qwen-Long2026.0468.9
AflowGenerator Backbone=GPT-4o-mini2026.0467.9
Self-ActGenerator Backbone=Qwen-Max2026.0467.8
AflowGenerator Backbone=GPT-4.1-nano2026.0466.5
Self-ActGenerator Backbone=Gemma-2-27B2026.0465.3
Self-ActGenerator Backbone=GPT-4o-mini2026.0462.4
Self-ActGenerator Backbone=GPT-4.1-nano2026.0459.8
ReActGenerator Backbone=Qwen-Max2026.0459.8
ReActGenerator Backbone=Gemma-2-27B2026.0457.8
ReActGenerator Backbone=Qwen-Long2026.0456.3
PiCABackbone=Qwen2.5-7B Instruct2026.0555.3
DirectGenerator Backbone=Qwen-Max2026.0453.3
TIPSBackbone=Qwen2.5-7B Instruct2026.0553.2
GraphRAG-RouterSystem=Our Method2026.0352.5
PiCABackbone=Qwen2.5-3B Instruct2026.0552.1
TIPSBackbone=Qwen2.5-3B Instruct2026.0551.8
DirectGenerator Backbone=Gemma-2-27B2026.0451.3
MT-PPOBackbone=Qwen2.5-7B Instruct2026.0550.6
Search-R1Backbone=Qwen2.5-7B Instruct2026.0549.6
Router-R1System=RL-based Agent2026.0349.5
ReActGenerator Backbone=GPT-4o-mini2026.0448.9
GraphSearchSystem=Training-free Agent2026.0348.3
MT-PPOBackbone=Qwen2.5-3B Instruct2026.0547.8
Search-o1System=Training-free Agent2026.0347.6
Search-R1System=RL-based Agent2026.0347.3
HippoRAG2System=RAG-Based2026.0346.5
IGPOBackbone=Qwen2.5-7B-Instruct2025.1046.4
Search-R1Backbone=Qwen2.5-3B Instruct2026.0545.7
ReActGenerator Backbone=GPT-4.1-nano2026.0445.4
LinearRAGSystem=RAG-Based2026.0343.6
DirectGenerator Backbone=Qwen-Long2026.0443.4
GPT-4o (2024-11-20)Version=2024-11-202025.0842.74
GraphRAGSystem=RAG-Based2026.0342.3
GSPOBackbone=Qwen2.5-7B-Instruct2025.1041.5
Graph-R1System=RL-based Agent2026.0341.5
RLOOBackbone=Qwen2.5-7B-Instruct2025.1040.7
GRPOBackbone=Qwen2.5-7B-Instruct2025.1040.3
Vanilla RAGSystem=RAG-Based2026.0340.1
CoTSystem=Basic LLM2026.0339.9
PPOBackbone=Qwen2.5-7B-Instruct2025.1038.7
RAPTORSystem=RAG-Based2026.0338.2
HyperGraphRAGSystem=RAG-Based2026.0337.5
DirectGenerator Backbone=GPT-4o-mini2026.0436.4
Reinforce++Backbone=Qwen2.5-7B-Instruct2025.1034.3
DirectGenerator Backbone=GPT-4.1-nano2026.0433.7
Llama + LoRRBackbone=Llama3.2-3B, Optimization=LoRR2025.0818.24
Llama + DPOBackbone=Llama3.2-3B, Optimization=DPO2025.0818.03
Qwen + LoRRBackbone=Qwen2.5-3B, Optimization=LoRR2025.0817.2
Qwen (Base)Backbone=Qwen2.5-3B, Mode=Base2025.0817.18
Qwen + DPOBackbone=Qwen2.5-3B, Optimization=DPO2025.0816.83
Direct InferSystem=Basic LLM2026.0316.5
Llama (Base)Backbone=Llama3.2-3B, Mode=Base2025.0814.75