List your company

Multi-Hop Search-augmented Question Answering on Bamboogle (success rate %)

68.4Success RateRole-Agent
5.27221.66138.0554.439May 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-Instruct2026.0668.4
No SkillBackbone=Qwen3-8B2026.0568.1
+ Base SkillBackbone=Qwen3-32B2026.0567.7
+ MASABackbone=Qwen3-32B2026.0566.1
No SkillBackbone=Qwen3-32B2026.0564.5
+ DS-AdapterBackbone=Qwen3-32B2026.0564.1
GiGPOType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0664.1
+ MASABackbone=Qwen3-8B2026.0562.5
+ MASABackbone=Qwen3-4B2026.0561.3
StepSearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0634.4
R1-InstructType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0629.3
Search_R1Type=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0626.4
+ Base SkillBackbone=Qwen3-14B2026.0515.3
+ DS-AdapterBackbone=Qwen3-4B2026.0512.9
+ DS-AdapterBackbone=Qwen3-14B2026.0512.5
Zero-SearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0611.1
No SkillBackbone=Qwen3-14B2026.0510.5
+ Base SkillBackbone=Qwen3-4B2026.0510.1
+ Base SkillBackbone=Qwen3-8B2026.0510.1
No SkillBackbone=Qwen3-4B2026.059.3
+ MASABackbone=Qwen3-14B2026.058.9
+ DS-AdapterBackbone=Qwen3-8B2026.057.7