Question Answering on NQ, TriviaQA, PopQA, HotpotQA, 2wiki, MuSiQue, Bamboogle
48.2Average QA ScoreIGRPO
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| IGRPOBackbone=Qwen2.5-7B-Instruct, Type=Tree-based2026.07 | 48.2 | — | — | — | — | — | — | — | — | — | — | |
| SAPOBackbone=Qwen2.5-7B-Instruct2026.06 | 47.8 | 48.4 | 62.9 | 46.7 | 45 | 45.2 | 18.3 | 46.4 | — | — | — | |
| GiGPOBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 47.3 | — | — | — | — | — | — | — | — | — | — | |
| IGPOBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 45.7 | — | — | — | — | — | — | — | — | — | — | |
| SKILLRLBackbone=Qwen2.5-7B-Instruct2026.06 | 45.5 | 45.9 | 61.2 | 45.9 | 43.9 | 39.9 | 17.2 | 45.6 | — | — | — | |
| IGRPOBackbone=Qwen2.5-3B-Instruct, Type=Tree-based2026.07 | 45.4 | — | — | — | — | — | — | — | — | — | — | |
| ZeroSearchBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 44.8 | — | — | — | — | — | — | — | — | — | — | |
| Skill0Backbone=Qwen2.5-7B-Instruct2026.06 | 44.4 | 42.7 | 63.6 | 45.3 | 40 | 38.3 | 16.4 | 66.9 | — | — | — | |
| Teacher-8BModel Family=Llama 3, Configuration=Teacher model2025.08 | 43.8 | 47.5 | 64.7 | 44.8 | 42.7 | 44.3 | 17.9 | 44.4 | — | — | — | |
| Tree-GRPOBackbone=Qwen2.5-7B-Instruct, Type=Tree-based2026.07 | 43.6 | — | — | — | — | — | — | — | — | — | — | |
| EvolveRBackbone=Qwen2.5-7B-Instruct2026.06 | 43.1 | 43.5 | 63.4 | 44.6 | 38.2 | 42 | 15.6 | 54.4 | — | — | — | |
| Search-R1Backbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 42.5 | — | — | — | — | — | — | — | — | — | — | |
| GiGPOBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 42.3 | — | — | — | — | — | — | — | — | — | — | |
| IGPOBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 41.4 | — | — | — | — | — | — | — | — | — | — | |
| SLIM2026.05 | 41 | 38.6 | 62.2 | 40 | 37.2 | 31.7 | 12.8 | 37.6 | 37.9 | 36.9 | — | |
| SLIM†variant=†2026.05 | 41 | 38.4 | 62.1 | 40.4 | 36.9 | 31.5 | 12.7 | 36 | 37.7 | 36.5 | — | |
| GRPOTeacher model=Qwen3-8B, Role=Teacher baseline2026.07 | 41 | 36.6 | 59.3 | 40.2 | 36.7 | 34.7 | 13.8 | 39.2 | — | — | — | |
| Tree-GRPOBackbone=Qwen2.5-3B-Instruct, Type=Tree-based2026.07 | 40.8 | — | — | — | — | — | — | — | — | — | — | |
| OPDTeacher model=Qwen3-4B-Instruct-2507, Student model=Qwen3-4B-Instruct-25072026.07 | 40.6 | 36.3 | 59.7 | 40.3 | 37.3 | 31.7 | 14.5 | 44 | — | — | — | |
| ReOPDTeacher model=Qwen3-4B-Instruct-2507, Student model=Qwen3-4B-Instruct-25072026.07 | 40.5 | 36.4 | 59.6 | 40.5 | 37.3 | 31.5 | 14 | 37.6 | — | — | — | |
| GRPOTeacher model=Qwen3-4B-Instruct-2507, Role=Teacher baseline2026.07 | 40.4 | 36.1 | 59.2 | 40.2 | 37.2 | 31.8 | 13.8 | 39.2 | — | — | — | |
| GRPO-SGBackbone=Qwen3-4B, Search Framework=VT-Search2025.10 | 40.18 | 48.23 | 64.11 | 46 | 36.6 | 30.89 | 10.26 | 41.6 | — | — | 14.1 | |
| Skill02026.05 | 39.3 | 37.9 | 59.5 | 38.6 | 32.7 | 31.9 | 10.3 | 32.8 | 35.3 | 34.6 | — | |
| ZeroSearchBackbone=Qwen2.5-7B-Instruct2026.06 | 39.1 | 43.6 | 61.8 | 51.5 | 34.6 | 35.2 | 18.4 | 27.8 | — | — | — | |
| OPDTeacher model=Qwen3-8B, Student model=Qwen3-4B-Instruct-25072026.07 | 39.1 | 35.5 | 57.3 | 39 | 35 | 31.2 | 12.8 | 37.6 | — | — | — | |
| ZeroSearchBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 39.1 | — | — | — | — | — | — | — | — | — | — | |
| ReOPDTeacher model=Qwen3-8B, Student model=Qwen3-4B-Instruct-25072026.07 | 39 | 34.9 | 57.5 | 39.1 | 34.6 | 31 | 11.9 | 37.6 | — | — | — | |
| DGPOModel Family=Llama 3, Distillation Configuration=8B → 1B2025.08 | 38.9 | 44.8 | 55.3 | 43.7 | 41.2 | 37.9 | 15.5 | 33.9 | — | — | — | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.06 | 38.5 | 39.3 | 61 | 39.7 | 37 | 40.1 | 14.6 | 36.8 | — | — | — | |
| SkillRL†variant=†2026.05 | 38.1 | 36.8 | 59.8 | 36.9 | 31.5 | 29.7 | 10.3 | 31.2 | 34.2 | 33.6 | — | |
| GRPO2026.05 | 37.5 | 35.9 | 57.8 | 36.5 | 30.8 | 30.1 | 9.2 | 30.4 | 33.4 | 32.8 | — | |
| Search-R1Backbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 37.4 | — | — | — | — | — | — | — | — | — | — | |
| ARBackbone=Qwen3-4B, Search Framework=VT-Search2025.10 | 36.11 | 49.55 | 61.45 | 36.98 | 38.19 | 23.95 | 8.25 | 34.4 | — | — | 2.6 | |
| BaseTeacher model=Qwen3-4B-Instruct-2507, Student model=Qwen3-4B-Instruct-25072026.07 | 35.6 | 28.4 | 55.5 | 30.6 | 30.8 | 33.6 | 8.4 | 35.8 | — | — | — | |
| 80/20Backbone=Qwen3-4B, Search Framework=VT-Search2025.10 | 35.4 | 47.94 | 62.67 | 40.27 | 37.52 | 25.66 | 6.51 | 27.2 | — | — | 0.5 | |
| GRPOBackbone=Qwen3-4B, Search Framework=VT-Search2025.10 | 35.21 | 46.48 | 59.84 | 40.39 | 35.54 | 28.87 | 10.26 | 27.2 | — | — | — | |
| KDModel Family=Llama 3, Distillation Configuration=8B → 1B2025.08 | 34.7 | 40.6 | 50.8 | 40.5 | 36.9 | 35.5 | 11.9 | 26.6 | — | — | — | |
| LoptiBackbone=Qwen3-4B, Search Framework=VT-Search2025.10 | 34.62 | 46.21 | 62.09 | 39.11 | 36.74 | 25.94 | 12.24 | 20 | — | — | 1.7 | |
| Cold StartTeacher model=Qwen3-4B-Instruct-2507, Student model=Qwen3-4B-Instruct-25072026.07 | 32.1 | 26.2 | 48.6 | 33.6 | 24.7 | 26.6 | 5.8 | 24.8 | — | — | — | |
| R1-InstructBackbone=Qwen2.5-3B-Instruct, Type=RL Training2026.07 | 29.8 | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-4BBackbone=Qwen3-4B, Search Framework=VT-Search2025.10 | 28.95 | 31.68 | 60 | 38.6 | 27.59 | 14.89 | 6.7 | 23.2 | — | — | — | |
| GRPO-SGBackbone=Qwen2.5-3B, Search Framework=VT-Search2025.10 | 27.29 | 41.8 | 52.3 | 39.1 | 24.5 | 20.3 | 5 | 8 | — | — | 97.2 | |
| R1-InstructBackbone=Qwen2.5-7B-Instruct, Type=RL Training2026.07 | 25.8 | — | — | — | — | — | — | — | — | — | — | |
| RAGBackbone=Qwen2.5-7B-Instruct2026.06 | 25.5 | 27.4 | 58.2 | 17.8 | 25.8 | 23.2 | 9.4 | 16.8 | — | — | — | |
| PPOModel Family=Llama 3, Distillation Configuration=8B → 1B2025.08 | 25 | 35.4 | 49.9 | 39.4 | 22.2 | 18.1 | 3.7 | 6.5 | — | — | — | |
| ARBackbone=Qwen2.5-3B, Search Framework=VT-Search2025.10 | 23.28 | 34.17 | 44.51 | 30.22 | 20.15 | 22.18 | 4.54 | 7.2 | — | — | 68.2 | |
| R1-InstructBackbone=Qwen2.5-7B-Instruct2026.06 | 22.4 | 21 | 44.9 | 17.1 | 20.8 | 27.5 | 6 | 19.2 | — | — | — | |
| Search-o1Backbone=Qwen2.5-7B-Instruct2026.06 | 22.1 | 19.4 | 40.6 | 11.4 | 17 | 27 | 8.6 | 30.4 | — | — | — | |
| LoptiBackbone=Qwen2.5-3B, Search Framework=VT-Search2025.10 | 21.62 | 40.22 | 40.19 | 25.6 | 16.53 | 18.22 | 4.18 | 6.4 | — | — | 56.2 | |
| 80/20Backbone=Qwen2.5-3B, Search Framework=VT-Search2025.10 | 18.6 | 25.95 | 38.55 | 17.28 | 18.54 | 14.3 | 3.61 | 12 | — | — | 34.4 | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.06 | 17.4 | 12.8 | 35.6 | 3.8 | 16.2 | 22.6 | 6.6 | 24 | — | — | — | |
| Qwen2.5Backbone=Qwen2.5-7B-Instruct2026.06 | 15.2 | 11.6 | 35.6 | 1.2 | 16.4 | 22.2 | 4.8 | 14.4 | — | — | — | |
| GRPOBackbone=Qwen2.5-3B, Search Framework=VT-Search2025.10 | 13.84 | 13.5 | 29.4 | 11.5 | 14.5 | 16.3 | 1.3 | 10.4 | — | — | — | |
| Student-1BModel Family=Llama 3, Configuration=Student baseline2025.08 | 3.9 | 5.2 | 8 | 4.4 | 2.7 | 4.2 | 0.1 | 2.4 | — | — | — | |
| Qwen2.5-3BBackbone=Qwen2.5-3B, Search Framework=VT-Search2025.10 | 0.64 | 0.5 | 1.2 | 0.8 | 0.5 | 1.5 | 0 | 0 | — | — | — | |
| StepSearchBackbone=Qwen2.5-7B-Instruct2026.06 | — | — | — | — | 38.6 | 36.6 | 22.6 | 40 | — | — | — |