List your company

General Question Answering on Natural Questions (NQ) (test val)

55.8EMDeepControl
0.1614.60529.0543.495Feb 2, 2026

Evaluation results

MethodLinks
DeepControlBackbone Model=Qwen2.5-7B2026.0255.847.9
DeepControlBackbone Model=Qwen2.5-3B2026.0253.341.1
Search-R1-baseBackbone Model=Qwen2.5-7B2026.024843.1
Search-R1-baseBackbone Model=Qwen2.5-3B2026.0240.630.3
Search-R1-instructBackbone Model=Qwen2.5-7B2026.0239.338.5
Rejection SamplingBackbone Model=Qwen2.5-7B2026.023634.8
RAGBackbone Model=Qwen2.5-7B2026.0234.930.4
RAGBackbone Model=Qwen2.5-3B2026.0234.827
Search-R1-instructBackbone Model=Qwen2.5-3B2026.0234.132.5
SFTBackbone Model=Qwen2.5-7B2026.0231.820.7
R1-baseBackbone Model=Qwen2.5-7B2026.0229.727.6
Rejection SamplingBackbone Model=Qwen2.5-3B2026.0229.426.5
R1-instructBackbone Model=Qwen2.5-7B2026.022727.1
SFTBackbone Model=Qwen2.5-3B2026.0224.917.6
Search-o1Backbone Model=Qwen2.5-3B2026.0223.825.5
R1-baseBackbone Model=Qwen2.5-3B2026.0222.622.9
IRCoTBackbone Model=Qwen2.5-7B2026.0222.423.9
R1-instructBackbone Model=Qwen2.5-3B2026.022122.4
Search-o1Backbone Model=Qwen2.5-7B2026.0215.120.6
Direct InferenceBackbone Model=Qwen2.5-7B2026.0213.418.1
IRCoTBackbone Model=Qwen2.5-3B2026.0211.118.1
Direct InferenceBackbone Model=Qwen2.5-3B2026.0210.613.4
CoTBackbone Model=Qwen2.5-7B2026.024.810.6
CoTBackbone Model=Qwen2.5-3B2026.022.31.5