Multi-Hop Question Answering on MuSiQue (Exact Match (EM))
22.6Exact Match (EM)StepSearch-Instruct
Evaluation Results
| Method | Links | |
|---|---|---|
| StepSearch-InstructTraining Pipeline=Pure RL, External Teacher=GPT-4o, Backbone=Qwen2.5-7B2026.05 | 22.6 | |
| SD-Search-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 22.2 | |
| MR-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 22.1 | |
| Thinker-InstructTraining Pipeline=Synthesize → SFT, External Teacher=Qwen2.5-72B, Backbone=Qwen2.5-7B2026.05 | 22.1 | |
| StepSearch-BaseTraining Pipeline=Pure RL, External Teacher=GPT-4o, Backbone=Qwen2.5-7B2026.05 | 21.6 | |
| SD-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 20.4 | |
| SkillRLEvaluation Protocol=Zero-shot2026.05 | 20.2 | |
| SKILLGRAPHEvaluation Protocol=Zero-shot2026.05 | 19.5 | |
| AutoRefine-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 18.7 | |
| ZeroSearchEvaluation Protocol=Zero-shot2026.05 | 18.4 | |
| ZEROSEARCH-instBackbone=Qwen-2.5-7B, Variant=Instruct2025.05 | 18.4 | |
| Search-R1-baseBackbone=Qwen-2.5-7B, Variant=Base2025.05 | 18.2 | |
| ZEROSEARCH-baseBackbone=Qwen-2.5-7B, Variant=Base2025.05 | 18 | |
| Search-R1-instBackbone=Qwen-2.5-7B, Variant=Instruct2025.05 | 17.4 | |
| Search-R1-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 17.3 | |
| EvolveREvaluation Protocol=Zero-shot2026.05 | 15.6 | |
| Search-R1Evaluation Protocol=Zero-shot2026.05 | 14.6 | |
| Search-R1-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 14.6 | |
| ZEROSEARCH-baseBackbone=LLaMA-3.2-3B, Variant=Base2025.05 | 13.8 | |
| Search-R1-baseBackbone=LLaMA-3.2-3B, Variant=Base2025.05 | 13.6 | |
| ZEROSEARCH-baseBackbone=Qwen-2.5-3B, Variant=Base2025.05 | 13 | |
| R1-instructBackbone=LLaMA-3.2-3B, Variant=Instruct2025.05 | 11.4 | |
| Search-R1-baseBackbone=Qwen-2.5-3B, Variant=Base2025.05 | 11.2 | |
| ZEROSEARCH-instBackbone=LLaMA-3.2-3B, Variant=Instruct2025.05 | 10.4 | |
| R1-baseBackbone=Qwen-2.5-7B, Variant=Base2025.05 | 9.8 | |
| ZEROSEARCH-instBackbone=Qwen-2.5-3B, Variant=Instruct2025.05 | 9.8 | |
| RAGBackbone=LLaMA-3.2-3B2025.05 | 9.6 | |
| RAGEvaluation Protocol=Zero-shot2026.05 | 9.4 | |
| RAGBackbone=Qwen-2.5-7B2025.05 | 9.4 | |
| Search-R1-instBackbone=LLaMA-3.2-3B, Variant=Instruct2025.05 | 8.8 | |
| Search-o1Evaluation Protocol=Zero-shot2026.05 | 8.6 | |
| Search-o1Backbone=Qwen-2.5-7B2025.05 | 8.6 | |
| R1-instructBackbone=Qwen-2.5-7B, Variant=Instruct2025.05 | 8.4 | |
| RAGBackbone=Qwen-2.5-3B2025.05 | 8.2 | |
| RAgentBackbone=Qwen-2.5-7B2025.05 | 7.6 | |
| Search-R1-instBackbone=Qwen-2.5-3B, Variant=Instruct2025.05 | 7.6 | |
| R1-baseBackbone=LLaMA-3.2-3B, Variant=Base2025.05 | 7 | |
| CoTEvaluation Protocol=Zero-shot2026.05 | 6.6 | |
| CoTBackbone=Qwen-2.5-7B2025.05 | 6.6 | |
| R1-baseBackbone=Qwen-2.5-3B, Variant=Base2025.05 | 6.4 | |
| Search-o1Backbone=LLaMA-3.2-3B2025.05 | 6 | |
| CoTBackbone=LLaMA-3.2-3B2025.05 | 5.8 | |
| RAgentBackbone=LLaMA-3.2-3B2025.05 | 5.6 | |
| Search-o1Backbone=Qwen-2.5-3B2025.05 | 5.2 | |
| Qwen2.5-7B-InstructEvaluation Protocol=Zero-shot2026.05 | 4.8 | |
| Direct AnswerBackbone=Qwen-2.5-7B2025.05 | 4.8 | |
| Direct AnswerBackbone=Qwen-2.5-3B2025.05 | 4.4 | |
| R1-instructBackbone=Qwen-2.5-3B, Variant=Instruct2025.05 | 4.4 | |
| CoTBackbone=Qwen-2.5-3B2025.05 | 3.6 | |
| RAgentBackbone=Qwen-2.5-3B2025.05 | 2.6 | |
| Direct AnswerBackbone=LLaMA-3.2-3B2025.05 | 2 |