Question Answering on Average (PopQA, HotpotQA, 2Wiki, MuSiQue, Bamboogle)
42.54EMQwen3-1.7B-SFT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-1.7B-SFTBackbone=GLM-5.1, Evaluation Protocol=mean@42026.07 | 42.54 | 55.03 | 63.2 | |
| DeepSeek-V4-FlashBackbone=GLM-5.1, Evaluation Protocol=mean@42026.07 | 42.13 | 54.39 | 63.15 | |
| Qwen3-1.7B-SFTBackbone=DeepSeek-V4-Flash, Evaluation Protocol=mean@42026.07 | 41.81 | 53.97 | 61.69 | |
| DeepSeek-V4-FlashBackbone=DeepSeek-V4-Flash, Evaluation Protocol=mean@42026.07 | 41.77 | 53.34 | 61.21 | |
| ProRAGMethod Category=Reinforcement Learning-based Methods2026.01 | 40.7 | 49.2 | — | |
| Qwen3-1.7BBackbone=DeepSeek-V4-Flash, Evaluation Protocol=mean@42026.07 | 39.18 | 50.72 | 57.38 | |
| Qwen3-1.7BBackbone=GLM-5.1, Evaluation Protocol=mean@42026.07 | 38.92 | 51.2 | 58.52 | |
| Search-R1Method Category=Reinforcement Learning-based Methods2026.01 | 38.5 | 46.7 | — | |
| ReasonRAGMethod Category=Reinforcement Learning-based Methods2026.01 | 34.5 | 42.3 | — | |
| HiPRAGMethod Category=Reinforcement Learning-based Methods2026.01 | 32.6 | 42.3 | — | |
| Search-o1Method Category=Advanced Methods2026.01 | 28.2 | 36 | — | |
| FLAREMethod Category=Advanced Methods2026.01 | 25.1 | 33.4 | — | |
| IRCoTMethod Category=Advanced Methods2026.01 | 21.6 | 30.4 | — | |
| Iter-RetGenMethod Category=Advanced Methods2026.01 | 16.1 | 20.5 | — | |
| Standard RAGMethod Category=Standard Methods2026.01 | 15.7 | 23.3 | — | |
| Naïve GenerationMethod Category=Standard Methods2026.01 | 8.8 | 15.4 | — |