List your company

Single-Hop Search-augmented Question Answering on PopQA

49.8Success RateRole-Agent
18.70426.77734.8542.923May 29, 2026May 30, 2026Jun 1, 2026Jun 3, 2026Jun 5, 2026Jun 7, 2026Jun 9, 2026

Evaluation results

MethodLinks
Role-AgentType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0649.8
Zero-SearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0644.8
GiGPOType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0642.4
+ DS-AdapterBackbone=Qwen3-4B2026.0541.8
+ MASABackbone=Qwen3-14B2026.0540.7
No SkillBackbone=Qwen3-14B2026.0540.6
+ DS-AdapterBackbone=Qwen3-32B2026.0540.6
+ MASABackbone=Qwen3-32B2026.0540
+ Base SkillBackbone=Qwen3-14B2026.0539.5
+ DS-AdapterBackbone=Qwen3-14B2026.0539.5
+ Base SkillBackbone=Qwen3-32B2026.0539.3
+ MASABackbone=Qwen3-8B2026.0539
+ MASABackbone=Qwen3-4B2026.0538.9
+ Base SkillBackbone=Qwen3-8B2026.0538.8
+ DS-AdapterBackbone=Qwen3-8B2026.0538.7
No SkillBackbone=Qwen3-32B2026.0538.3
+ Base SkillBackbone=Qwen3-4B2026.0538.2
Search_R1Type=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0637.8
No SkillBackbone=Qwen3-4B2026.0537.2
No SkillBackbone=Qwen3-8B2026.0530.3
R1-InstructType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0619.9