List your company

Single-Hop Question Answering on Single-Hop Suite (NQ, TriviaQA, PopQA) (test)

49.64NQ AccuracyAKBE (Ours)
2.52814.75926.9939.221May 26, 2026

Evaluation results

MethodLinks
AKBE (Ours)Backbone Model=Qwen2.5-7B2026.0549.6464.6347.9154.59154.59
OTC-POBackbone Model=Qwen2.5-7B2026.0548.7263.4647.6453.89153.89
β-GRPOBackbone Model=Qwen2.5-7B2026.0547.8964.2148.1854.351.0352.76
AKBE (Ours)Backbone Model=Qwen3-4B2026.0547.3365.1850.855.91.1648.19
Offline AKBEBackbone Model=Qwen2.5-7B2026.0547.1563.9248.0554.081.0153.54
β-GRPOBackbone Model=Qwen3-4B2026.054764.7349.9555.31.5336.15
R1-SearcherBackbone Model=Qwen3-4B2026.0546.8564.5249.4354.951.4837.13
Offline AKBEBackbone Model=Qwen3-4B2026.0546.5264.5549.7255.061.2245.13
Search-R1Backbone Model=Qwen3-4B2026.0545.9865.1749.1854.971.4138.99
OTC-POBackbone Model=Qwen3-4B2026.0545.3164.3549.8654.91.0253.83
R1-SearcherBackbone Model=Qwen2.5-7B2026.0545.1263.2849.0554.071.0850.06
Search-R1Backbone Model=Qwen2.5-7B2026.0544.3563.6548.753.961.0551.39
Search-o1Backbone Model=Qwen3-4B2026.0533.8557.4240.2746.111.7226.81
ReActBackbone Model=Qwen3-4B2026.0526.7553.5335.3441.312.2718.19
Search-o1Backbone Model=Qwen2.5-7B2026.0515.8732.4522.1825.371.2819.82
ReActBackbone Model=Qwen2.5-7B2026.054.3410.679.329.231.366.79