Multi-hop QA Retrieval on 2WikiMultihopQA (test)
97.2R@5CoopRAG
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CoopRAGLLM=Gemma2-27B, Single-step retrieval=true2025.12 | 97.2 | 80.8 | — | — | — | — | — | — | |
| CoopRAGBackbone LLM=Gemma2-27B2025.12 | 97.2 | 80.8 | — | — | — | — | — | — | |
| CoopRAGLLM=Gemma2-9B, Single-step retrieval=true2025.12 | 96.7 | 80.1 | — | — | — | — | — | — | |
| CoopRAGBackbone LLM=Gemma2-9B2025.12 | 96.7 | 80.1 | — | — | — | — | — | — | |
| CoopRAGLLM=GPT-4o-mini, Single-step retrieval=true2025.12 | 96.6 | 80.4 | — | — | — | — | — | — | |
| CoopRAGBackbone LLM=GPT-4o-mini2025.12 | 96.6 | 80.4 | — | — | — | — | — | — | |
| CoopRAGLLM=Llama3.3-70B, Single-step retrieval=true2025.12 | 96.3 | 80.6 | — | — | — | — | — | — | |
| CoopRAGBackbone LLM=Llama3.3-70B2025.12 | 96.3 | 80.6 | — | — | — | — | — | — | |
| BRIDGERAGCondition=C (+bridge cond.)2026.04 | 95.27 | — | — | — | — | — | 1.2 | — | |
| BRIDGERAGCondition=B (+2-way judge)2026.04 | 94.4 | — | — | — | — | — | 0.3 | — | |
| PROPRAG2026.04 | 94.1 | — | — | — | — | — | — | — | |
| HippoRAG2LLM=Llama3.3-70B, Single-step retrieval=true2025.12 | 90.4 | 76.2 | — | — | — | — | — | — | |
| HippoRAGBackbone LLM=Llama3.3-70B2025.12 | 90.4 | 71.9 | — | — | — | — | — | — | |
| HippoRAG2Backbone LLM=Llama3.3-70B2025.12 | 90.4 | 76.2 | — | — | — | — | — | — | |
| HIPPORAG22026.04 | 90.4 | — | — | — | — | — | 3.7 | — | |
| HippoRAG2LLM=GPT-4o-mini, Single-step retrieval=true2025.12 | 90.2 | 74.6 | — | — | — | — | — | — | |
| HippoRAG2Backbone LLM=GPT-4o-mini2025.12 | 90.2 | 74.6 | — | — | — | — | — | — | |
| BRIDGERAGCondition=A (SVO only)2026.04 | 89.1 | — | — | — | — | — | 5 | — | |
| HippoRAGBackbone LLM=GPT-4o-mini2025.12 | 87 | 68.4 | — | — | — | — | — | — | |
| NV-Embed-v2Parameters=7B2025.12 | 76.5 | 67.1 | — | — | — | — | — | — | |
| GritLM-7B2025.12 | 76 | 67.3 | — | — | — | — | — | — | |
| GTE-Qwen2-7B-Instruct2025.12 | 74.8 | 66.7 | — | — | — | — | — | — | |
| HopRAGLLM=GPT-4o-mini, Single-step retrieval=true2025.12 | 70.1 | 61.7 | — | — | — | — | — | — | |
| HopRAGBackbone LLM=GPT-4o-mini2025.12 | 70.1 | 61.7 | — | — | — | — | — | — | |
| ColBERTv22025.12 | 68.2 | 59.2 | — | — | — | — | — | — | |
| GTRBackbone=T5-base2025.12 | 67.9 | 60.2 | — | — | — | — | — | — | |
| SiReRAGLLM=GPT-4o-mini, Single-step retrieval=true2025.12 | 67.6 | 60.6 | — | — | — | — | — | — | |
| SiReRAGBackbone LLM=GPT-4o-mini2025.12 | 67.6 | 60.6 | — | — | — | — | — | — | |
| RAPTORBackbone LLM=Llama3.3-70B2025.12 | 66.2 | 58.3 | — | — | — | — | — | — | |
| RAPTORBackbone LLM=GPT-4o-mini2025.12 | 66 | 58.4 | — | — | — | — | — | — | |
| BM252025.12 | 65.3 | 55.3 | — | — | — | — | — | — | |
| PropositionBase Retriever=ColBERTv22025.12 | 64.9 | 55.9 | — | — | — | — | — | — | |
| Contriever2025.12 | 57.5 | 46.6 | — | — | — | — | — | — | |
| Auto-RAGModel Parameters=7B2026.05 | — | — | — | — | — | — | — | 66.05 | |
| BGE-largeModel Parameters=326M2026.05 | — | — | — | — | — | — | — | 58.43 | |
| BM252026.05 | — | — | — | — | — | — | — | 49.8 | |
| Contriever2026.05 | — | — | — | — | — | — | — | 51.25 | |
| GritHopperModel Parameters=7B2026.05 | — | — | — | — | — | — | — | 59.97 | |
| IRCoTModel Parameters=70B2026.05 | — | — | — | — | — | — | — | 65.12 | |
| ITER-RETGENModel Parameters=70B2026.05 | — | — | — | — | — | — | — | 59.96 | |
| MDRModel Parameters=110M2026.05 | — | — | — | — | — | — | — | 53.91 | |
| R3-RAGModel Parameters=8B2026.05 | — | — | — | — | — | — | — | 82.82 | |
| RT-RAGModel Parameters=8B2026.05 | — | — | — | — | — | — | — | 65.14 | |
| SEALModel Parameters=406M2026.05 | — | — | — | — | — | — | — | 48.32 | |
| SelfaskModel Parameters=70B2026.05 | — | — | — | — | — | — | — | 47.07 | |
| Thermovariant=tune winner, α=0.4, dk=202026.03 | — | — | 43.5 | 26 | 11 | 0.02 | — | — | |
| Thermovariant=best recall, α=0.4, dk=502026.03 | — | — | 44.3 | 24 | 7 | 0.003 | — | — | |
| Thermovariant=safest, α=0.5, dk=502026.03 | — | — | 43.1 | 15 | 1 | 0.001 | — | — | |
| ThinkGRModel Parameters=8B2026.05 | — | — | — | — | — | — | — | 93.19 | |
| True RRF†2026.03 | — | — | 41.2 | 15 | 6 | 0.078 | — | — | |
| Vector only2026.03 | — | — | 37.8 | — | — | — | — | — | |
| w/o RLModel Parameters=8B2026.05 | — | — | — | — | — | — | — | 92.03 | |
| w/o SFTModel Parameters=8B2026.05 | — | — | — | — | — | — | — | 57.7 | |
| w/o ThoughtModel Parameters=8B2026.05 | — | — | — | — | — | — | — | 93.01 |