Multi-hop Question Answering on MuSiQue (val)
810Avg Token BudgetSELFCEST
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SELFCESTMechanism=Spawn clones with retrieval key, Training=End-to-end reinforcement learning2026.02 | 810 | 51 | — | — | — | |
| Base + toolsAugmentation=Information retrieval tool, Protocol=Agentic RAG pipeline2026.02 | 860 | 14 | — | — | — | |
| CRAGBackbone=Llama-3.1-8B2026.07 | 1,477.9 | — | 19.8 | — | 10.76 | |
| Adaptive-RAGBackbone=Llama-3.1-8B2026.07 | 1,749.8 | — | 17.19 | — | 9.52 | |
| Adaptive-RAGBackbone=GPT-4o-mini2026.07 | 1,789.1 | — | 31.72 | — | 19.28 | |
| IRCoTBackbone=GPT-4o-mini2026.07 | 1,939.6 | — | 38.7 | — | 24.53 | |
| CRAGBackbone=GPT-4o-mini2026.07 | 2,020.7 | — | 35.78 | — | 21.51 | |
| IRCoTBackbone=Llama-3.1-8B2026.07 | 2,049.1 | — | 21.26 | — | 13.16 | |
| IRCoTBackbone=Qwen2.5-7B2026.07 | 2,109.6 | — | 28.89 | — | 18.62 | |
| Adaptive-RAGBackbone=Qwen2.5-7B2026.07 | 2,160.4 | — | 25.79 | — | 16.47 | |
| Base Qwen3-4BContext=Full context2026.02 | 2,417 | 35 | — | — | — | |
| CRAGBackbone=Qwen2.5-7B2026.07 | 2,479.5 | — | 27.4 | — | 16.76 | |
| DynaKRAGBackbone=GPT-4o-mini2026.07 | 2,499.9 | — | 39.77 | — | 26.35 | |
| Fixed-KBackbone=GPT-4o-mini2026.07 | 2,763.6 | — | 32.79 | — | 18.99 | |
| Fixed-KBackbone=Llama-3.1-8B2026.07 | 2,814.2 | — | 14.61 | — | 6.95 | |
| Fixed-KBackbone=Qwen2.5-7B2026.07 | 2,918.3 | — | 25.53 | — | 15.18 | |
| CoRAG-s3Backbone=Llama-3.1-8B2026.07 | 2,945.1 | — | 16.9 | — | 9.56 | |
| CoRAG-s3Backbone=GPT-4o-mini2026.07 | 3,046 | — | 38.72 | — | 24.82 | |
| DynaKRAGBackbone=Qwen2.5-7B2026.07 | 3,082.3 | — | 30.61 | — | 19.86 | |
| S2G-RAGBackbone=GPT-4o-mini2026.07 | 3,091.2 | — | 36.91 | — | 22.8 | |
| DynaKRAGBackbone=Llama-3.1-8B2026.07 | 3,254.6 | — | 23.91 | — | 15.14 | |
| CoRAG-s3Backbone=Qwen2.5-7B2026.07 | 3,263.6 | — | 28.48 | — | 18.58 | |
| PAR2-RAGBackbone=Llama-3.1-8B2026.07 | 3,333.6 | — | 16.74 | — | 8.81 | |
| PAR2-RAGBackbone=GPT-4o-mini2026.07 | 3,477.4 | — | 32.41 | — | 18.87 | |
| PAR2-RAGBackbone=Qwen2.5-7B2026.07 | 3,755.6 | — | 24.8 | — | 14.48 | |
| S2G-RAGBackbone=Llama-3.1-8B2026.07 | 3,848.5 | — | 19.73 | — | 11.46 | |
| S2G-RAGBackbone=Qwen2.5-7B2026.07 | 3,886.8 | — | 29.99 | — | 19.36 | |
| Self-Ask+SearchBackbone=Llama-3.1-8B2026.07 | 4,000.1 | — | 16.85 | — | 9.35 | |
| Self-Ask+SearchBackbone=GPT-4o-mini2026.07 | 4,570 | — | 38.76 | — | 24.49 | |
| Self-Ask+SearchBackbone=Qwen2.5-7B2026.07 | 4,761.8 | — | 29.98 | — | 19.74 | |
| Qwen3-32B-BaseSetting=Zero-shot, Mode=Closed-book2026.05 | — | 3.13 | 13.14 | — | — | |
| Qwen3-32B-DS-SFTSetting=Zero-shot, Mode=Closed-book, Teacher LLM=DeepSeek2026.05 | — | 3.4 | 13.61 | 19.7 | — | |
| Qwen3-32B-GE-SFTSetting=Zero-shot, Mode=Closed-book, Teacher LLM=Gemini2026.05 | — | 3.2 | 13.65 | 13.6 | — |