Multi-hop Question Answering on HotpotQA (LM, F1, Acc)
83.2F1 ScoreWORC+AC
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| WORC+AC2026.04 | 83.2 | — | — | — | |
| AoT2026.04 | 80.6 | — | — | — | |
| AgentChain2026.04 | 75.1 | — | — | — | |
| AFlow2026.04 | 73.5 | — | — | — | |
| Self-Refine2026.04 | 68.3 | — | — | — | |
| CoT2026.04 | 67.2 | — | — | — | |
| FoTn (sampling size)=82026.04 | 66.7 | — | — | — | |
| CoT-SCn (sampling size)=52026.04 | 66.2 | — | — | — | |
| Analogical Prompting2026.04 | 64.7 | — | — | — | |
| GoldRouter (Multi-RAG-Agent)Models=Qwen-Plus2025.01 | 63.99 | 58.17 | 81.01 | — | |
| EfficientRAG (Single-RAG-Agent)Models=Qwen-Plus2025.01 | 63.01 | 57.41 | 76.55 | — | |
| GoldRouter (Multi-RAG-Agent)Models=LLaMA-3.1-8B2025.01 | 56.26 | 50.6 | 64.94 | — | |
| EfficientRAG (Single-RAG-Agent)Models=LLaMA-3.1-8B2025.01 | 52.96 | 47.38 | 61.17 | — | |
| CoT (Without RAG)Models=Qwen-Plus2025.01 | 43.82 | 39.68 | 54.58 | — | |
| CoT (Without RAG)Models=LLaMA-3.1-8B2025.01 | 36.02 | 30.96 | 44.23 | — | |
| OracleRetrieval Type=Oracle, Generator=Qwen3-VL-8B-Instruct2025.04 | 34.72 | — | — | 26.35 | |
| UniversalRAG (Qwen3-VL-2B-Instruct)Router Type=Trained, Router Model=Qwen3-VL-2B-Instruct, Generator=Qwen3-VL-8B-Instruct2025.04 | 34.61 | — | — | 26.1 | |
| NaïveGenerator=Qwen3-VL-8B-Instruct2025.04 | 29.53 | — | — | 21.1 |