List your company

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

17.8Success RateRole-Agent
5.1128.40611.714.994May 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-Instruct2026.0617.8
StepSearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0617.4
GiGPOType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0612.6
+ MASABackbone=Qwen3-32B2026.0511.8
+ Base SkillBackbone=Qwen3-32B2026.0511.7
+ DS-AdapterBackbone=Qwen3-32B2026.0511.6
+ Base SkillBackbone=Qwen3-14B2026.0511.4
Search_R1Type=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0610.3
+ MASABackbone=Qwen3-8B2026.0510
Zero-SearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.069.8
+ MASABackbone=Qwen3-14B2026.059.7
+ MASABackbone=Qwen3-4B2026.059.4
+ DS-AdapterBackbone=Qwen3-4B2026.059.3
+ DS-AdapterBackbone=Qwen3-14B2026.059.2
No SkillBackbone=Qwen3-32B2026.058.6
+ Base SkillBackbone=Qwen3-4B2026.057.8
No SkillBackbone=Qwen3-14B2026.057.6
R1-InstructType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.067.2
No SkillBackbone=Qwen3-8B2026.056.7
No SkillBackbone=Qwen3-4B2026.056.4
+ Base SkillBackbone=Qwen3-8B2026.056.2
+ DS-AdapterBackbone=Qwen3-8B2026.055.6