List your company

Multi-Hop Search-augmented Question Answering on 2Wiki (success rate %)

45.2Success RateRole-Agent
21.90427.9523440.048May 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-Instruct2026.0645.2
GiGPOType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0637
+ MASABackbone=Qwen3-32B2026.0535.6
+ DS-AdapterBackbone=Qwen3-32B2026.0532
StepSearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0632
Search_R1Type=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0631.9
No SkillBackbone=Qwen3-8B2026.0530.6
+ Base SkillBackbone=Qwen3-14B2026.0530.3
+ MASABackbone=Qwen3-14B2026.0530
Zero-SearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0630
No SkillBackbone=Qwen3-32B2026.0529.3
R1-InstructType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0629.2
+ DS-AdapterBackbone=Qwen3-14B2026.0528.5
+ MASABackbone=Qwen3-4B2026.0527
No SkillBackbone=Qwen3-14B2026.0526.8
+ Base SkillBackbone=Qwen3-32B2026.0526
+ Base SkillBackbone=Qwen3-8B2026.0525.9
+ MASABackbone=Qwen3-8B2026.0525.7
+ Base SkillBackbone=Qwen3-4B2026.0524.4
+ DS-AdapterBackbone=Qwen3-4B2026.0523.9
+ DS-AdapterBackbone=Qwen3-8B2026.0522.9
No SkillBackbone=Qwen3-4B2026.0522.8