Multi-hop Question Answering on MuSiQue (test) (Cost and Repair Metrics)
4,987Token CostDR-RAG
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DR-RAGBackbone=Qwen3-8B, Baseline=Search-o1 [27]2026.04 | 4,987 | 20.8 | 10 | 10 | 10 | |
| DR-RAGBackbone=Qwen3-8B, Baseline=ReAct [45]2026.04 | 5,261 | 26.1 | 10 | 9.1 | 9.1 | |
| DR-RAGBackbone=Qwen3-8B, Baseline=Search-R1 [21]2026.04 | 5,395 | 20.6 | 10 | 10.5 | 10.2 | |
| DR-RAGBackbone=Qwen3-4B, Baseline=ReAct [45]2026.04 | 5,432 | 23.1 | 8.4 | 8.5 | 9.1 | |
| DR-RAGBackbone=Qwen3-4B, Baseline=Search-R1 [21]2026.04 | 5,728 | 17.9 | 5.6 | 6.7 | 6.7 | |
| DR-RAGBackbone=Qwen3-4B, Baseline=Search-o1 [27]2026.04 | 5,732 | 17.1 | 9.2 | 10.5 | 10.4 | |
| Step-wiseBackbone=Qwen3-4B, Baseline=Search-R1 [21]2026.04 | 6,162 | 5.7 | 2.6 | 3.7 | 3.9 | |
| DR-RAGBackbone=LLaMA-3.1-8B-Instruct, Baseline=Search-o1 [27]2026.04 | 6,609 | 13.9 | 5.6 | 4.8 | 4.6 | |
| RerunBackbone=Qwen3-8B, Baseline=Search-R1 [21]2026.04 | 6,632 | 16.7 | 6.6 | 8.2 | 8.1 | |
| RerunBackbone=Qwen3-8B, Baseline=Search-o1 [27]2026.04 | 6,809 | 7.3 | 4.6 | 3.6 | 3.6 | |
| Step-wiseBackbone=Qwen3-8B, Baseline=Search-R1 [21]2026.04 | 7,057 | 7.3 | 1.4 | 2 | 2.2 | |
| DR-RAGBackbone=LLaMA-3.1-8B-Instruct, Baseline=ReAct [45]2026.04 | 7,196 | 20.1 | 7 | 5.4 | 5.2 | |
| DR-RAGBackbone=LLaMA-3.1-8B-Instruct, Baseline=Search-R1 [21]2026.04 | 7,268 | 12.8 | 5.2 | 5.9 | 5.9 | |
| Step-wiseBackbone=Qwen3-8B, Baseline=Search-o1 [27]2026.04 | 7,320 | 7.7 | 3 | 3.4 | 3 | |
| RAG-Critic [8]Backbone=LLaMA-3.1-8B-Instruct, Baseline=Search-R1 [21]2026.04 | 7,519 | 2.6 | 1 | 0.5 | 0.6 | |
| Step-wiseBackbone=Qwen3-4B, Baseline=Search-o1 [27]2026.04 | 7,664 | 5.3 | 2.6 | 2.3 | 1.9 | |
| RerunBackbone=Qwen3-4B, Baseline=Search-R1 [21]2026.04 | 7,874 | 13.5 | 4.2 | 5.9 | 6.3 | |
| RerunBackbone=Qwen3-4B, Baseline=Search-o1 [27]2026.04 | 7,881 | 7.7 | 4 | 4.4 | 4.2 | |
| RerunBackbone=Qwen3-8B, Baseline=ReAct [45]2026.04 | 8,113 | 15.7 | 6 | 7.4 | 7.3 | |
| RAG-Critic [8]Backbone=LLaMA-3.1-8B-Instruct, Baseline=ReAct [45]2026.04 | 8,128 | 10.7 | 4 | 2.2 | 2.3 | |
| RerunBackbone=Qwen3-4B, Baseline=ReAct [45]2026.04 | 8,773 | 15 | 5 | 5.8 | 5.6 | |
| RAG-Critic [8]Backbone=LLaMA-3.1-8B-Instruct, Baseline=Search-o1 [27]2026.04 | 9,486 | 4.6 | 1.8 | 1.6 | 1.7 | |
| RerunBackbone=LLaMA-3.1-8B-Instruct, Baseline=Search-R1 [21]2026.04 | 9,502 | 6.7 | 3.4 | 4.6 | 4.9 | |
| Step-wiseBackbone=Qwen3-4B, Baseline=ReAct [45]2026.04 | 9,871 | 9.8 | 4.2 | 3.3 | 3.3 | |
| Step-wiseBackbone=Qwen3-8B, Baseline=ReAct [45]2026.04 | 9,927 | 11.5 | 3.6 | 3.8 | 3.7 | |
| RerunBackbone=LLaMA-3.1-8B-Instruct, Baseline=Search-o1 [27]2026.04 | 12,065 | 5.6 | 2.8 | 3.6 | 3.4 | |
| Step-wiseBackbone=LLaMA-3.1-8B-Instruct, Baseline=Search-R1 [21]2026.04 | 13,162 | 5.3 | 2.4 | 2.8 | 3.3 | |
| RAG-Critic [8]Backbone=Qwen3-4B, Baseline=ReAct [45]2026.04 | 14,186 | 12.1 | 4.8 | 4.3 | 4.4 | |
| RerunBackbone=LLaMA-3.1-8B-Instruct, Baseline=ReAct [45]2026.04 | 14,787 | 14 | 4.2 | 4.5 | 4.3 | |
| RAG-Critic [8]Backbone=Qwen3-8B, Baseline=ReAct [45]2026.04 | 15,610 | 17.9 | 8.6 | 8.1 | 8.5 | |
| RAG-Critic [8]Backbone=Qwen3-8B, Baseline=Search-R1 [21]2026.04 | 16,099 | 11.9 | 7.8 | 8.7 | 9.5 | |
| Step-wiseBackbone=LLaMA-3.1-8B-Instruct, Baseline=Search-o1 [27]2026.04 | 16,477 | 6.3 | 4 | 4.4 | 4 | |
| RAG-Critic [8]Backbone=Qwen3-4B, Baseline=Search-R1 [21]2026.04 | 17,685 | 6.1 | 1.2 | 1.1 | 1.7 | |
| RAG-Critic [8]Backbone=Qwen3-8B, Baseline=Search-o1 [27]2026.04 | 19,763 | 9.7 | 7.6 | 8.5 | 8.8 | |
| RAG-Critic [8]Backbone=Qwen3-4B, Baseline=Search-o1 [27]2026.04 | 19,770 | 5.8 | 3 | 3.7 | 3.9 | |
| Step-wiseBackbone=LLaMA-3.1-8B-Instruct, Baseline=ReAct [45]2026.04 | 21,239 | 13 | 4.6 | 4.7 | 4.8 |