General Question Answering on NQ, TriviaQA, and PopQA
58.2NQ AccuracyQwen2.5-32B-Instruct + Proposed Method
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen2.5-32B-Instruct + Proposed MethodBackbone=Qwen2.5-32B-Instruct, Training=Proposed Method2026.05 | 58.2 | 81.8 | 58.2 | — | — | |
| Qwen2.5-32B-Instruct + SSPBackbone=Qwen2.5-32B-Instruct, Training=SSP2026.05 | 56.4 | 78.2 | 55.6 | — | — | |
| Qwen2.5-32B-InstructBackbone=Qwen2.5-32B-Instruct2026.05 | 56.2 | 77.6 | 55 | — | — | |
| Qwen2.5-14B-Instruct + Proposed MethodBackbone=Qwen2.5-14B-Instruct, Training=Proposed Method2026.05 | 55.8 | 80 | 56.8 | — | — | |
| Qwen2.5-14B-Instruct + SSPBackbone=Qwen2.5-14B-Instruct, Training=SSP2026.05 | 55.2 | 78 | 56 | — | — | |
| Qwen2.5-14B-InstructBackbone=Qwen2.5-14B-Instruct2026.05 | 53.2 | 76.2 | 55 | — | — | |
| Search-R1-GRPO + LLDSBackbone=Qwen2.5-7b-Ins2025.12 | 51.8 | 67.6 | 48.6 | 56 | 48.3 | |
| Search-R1-GRPO + LLDSBackbone=Qwen2.5-7b-Base2025.12 | 51.2 | 67.2 | 48 | 55.5 | 48 | |
| GRPO-LLDS-MABackbone=Llama3.2-3b-Base2025.12 | 49.3 | 64 | 48 | 53.8 | — | |
| GRPO-LLDSBackbone=Llama3.2-3b-Base2025.12 | 48.6 | 64.3 | 46.7 | 53.2 | — | |
| Tree-GRPOBackbone=Qwen2.5-7B2025.12 | 48.1 | 63.3 | 45.2 | 52.2 | 43.9 | |
| Search-R1-PPO-BaseBackbone=Qwen2.5-7B-Base2026.02 | 48 | 63.8 | 45.7 | 52.5 | 43.1 | |
| Search-R1-PPO-BaseBackbone=Qwen2.5-7B, Alignment=Base2025.12 | 48 | 63.8 | 45.7 | 52.5 | 43.1 | |
| DARTBackbone=Qwen2.5-7B-Base2026.02 | 47.2 | 63.9 | 50.7 | 53.9 | 41.6 | |
| DARTBackbone=Qwen2.5-7B-Instruct2026.02 | 46.7 | 64.2 | 50.5 | 53.8 | 42 | |
| ZeroSearch-InsBackbone=Qwen2.5-7B, Alignment=Instruct2025.12 | 43.6 | 65.2 | 48.8 | 52.5 | 39.1 | |
| Search-R1-GRPOBackbone=Qwen2.5-7B-Instruct2026.02 | 42.9 | 62.3 | 42.7 | 49.3 | 39.6 | |
| Search-R1-GRPOBackbone=Qwen2.5-7b-Ins2025.12 | 42.9 | 62.3 | 42.7 | 49.3 | 39.6 | |
| ZeroSearch-BaseBackbone=Qwen2.5-7B, Alignment=Base2025.12 | 42.4 | 66.4 | 60.4 | 56.4 | 40.9 | |
| Search-R1-GRPOBackbone=Qwen2.5-7B-Base2026.02 | 39.5 | 56 | 38.8 | 44.8 | 35 | |
| Search-R1-GRPOBackbone=Qwen2.5-7b-Base2025.12 | 39.5 | 56 | 38.8 | 44.8 | 35 | |
| Search-R1-PPO-InsBackbone=Qwen2.5-7B-Instruct2026.02 | 39.3 | 61 | 39.7 | 46.7 | 38.5 | |
| Search-R1-PPO-InsBackbone=Qwen2.5-7B, Alignment=Instruct2025.12 | 39.3 | 61 | 39.7 | 46.7 | 38.5 | |
| StepSearchBackbone=Qwen2.5-7B2025.12 | 36.4 | 56.5 | 39 | 44 | 38.2 | |
| Rejection SamplingBackbone=Qwen2.5-7B2026.02 | 36 | 59.2 | 38 | 44.4 | 34.8 | |
| Rejection SamplingBackbone=Qwen2.5-7B2025.12 | 36 | 59.2 | 38 | 44.4 | 34.8 | |
| RAGBackbone=Qwen2.5-7B2026.02 | 34.9 | 58.5 | 39.2 | 44.2 | 30.4 | |
| RAGBackbone=Qwen2.5-7B2025.12 | 34.9 | 58.5 | 39.2 | 44.2 | 30.4 | |
| SFTBackbone=Qwen2.5-7B2026.02 | 31.8 | 35.4 | 12.1 | 26.4 | 20.7 | |
| R1-baseBackbone=Qwen2.5-7B-Base2026.02 | 29.7 | 53.9 | 19.9 | 34.5 | 26.2 | |
| R1-baseBackbone=Qwen2.5-7B, Alignment=Base2025.12 | 29.7 | 53.9 | 19.9 | 34.5 | 26.2 | |
| R1-instructBackbone=Qwen2.5-7B-Instruct2026.02 | 27 | 53.7 | 19.9 | 33.5 | 27.1 | |
| R1-instructBackbone=Qwen2.5-7B, Alignment=Instruct2025.12 | 27 | 53.7 | 19.9 | 33.5 | 27.1 | |
| GRPOBackbone=Llama3.2-3b-Base2025.12 | 23.3 | 46.1 | 22 | 30.5 | — | |
| IRCoTBackbone=Qwen2.5-7B2026.02 | 22.4 | 47.8 | 30.1 | 33.4 | 23.9 | |
| Search-o1Backbone=Qwen2.5-7B2026.02 | 15.1 | 44.3 | 13.1 | 24.2 | 20.6 | |
| Search-o1Backbone=Qwen2.5-7B2025.12 | 15.1 | 44.3 | 13.1 | 24.2 | 20.6 | |
| Direct InferenceBackbone=Qwen2.5-7B2026.02 | 13.4 | 40.8 | 14 | 22.7 | 18.1 | |
| Direct InferenceBackbone=Qwen2.5-7B2025.12 | 13.4 | 40.8 | 14 | 22.7 | 18.1 | |
| CoTBackbone=Qwen2.5-7B2026.02 | 4.8 | 18.5 | 5.4 | 9.6 | 10.6 |