List your company

Multi-hop Document Retrieval on HotpotQA (test)

91.03Recall@KGritHopper
22.358840.186958.01575.8431Mar 2, 2025May 15, 2025Jul 28, 2025Oct 10, 2025Dec 23, 2025Mar 7, 2026May 21, 2026

Evaluation results

MethodLinks
GritHopperModel Parameters=7B2026.0591.03———
MDRModel Parameters=110M2026.0587.57———
IRCoTK=7.92025.0377———
ThinkGRModel Parameters=8B2026.0576.09———
IterRetGenK=6.02025.0373.2———
SelfAskK=28.62025.0371.1———
w/o ThoughtModel Parameters=8B2026.0569.86———
w/o RLModel Parameters=8B2026.0567.66———
Vanilla RAGK=8.02025.0364.1———
RT-RAGModel Parameters=8B2026.0562.48———
ITER-RETGENModel Parameters=70B2026.0561.94———
L-RAGK=8.02025.0361.3———
BGE-largeModel Parameters=326M2026.0560.48———
R3-RAGModel Parameters=8B2026.0558.56———
SEALModel Parameters=406M2026.0556.15———
IRCoTModel Parameters=70B2026.0555.79———
Auto-RAGModel Parameters=7B2026.0554.52———
Vanilla RAGK=4.02025.0354.3———
Contriever2026.0550.35———
BM252026.0546.39———
L-RAGK=4.02025.0344.9———
SelfaskModel Parameters=70B2026.0544.4———
w/o SFTModel Parameters=8B2026.0535.76———
HyDEK=6.02025.0325———
BGE-Large-EN-v1.5Method Type=Large Embedding Model2026.06—7688.8—
BM25Method Type=Simple Baseline, Source=Literature2026.06—57.374.8—
BRIDGERAGCondition=A (SVO only)2026.04——97.20.2
BRIDGERAGCondition=C (+bridge cond.)2026.04——98.751.35
ContrieverMethod Type=Simple Baseline, Source=Literature2026.06—58.475.3—
CoopRAGLLM=Gemma2-9B, Single-step retrieval=true2025.12—87.995.6—
CoopRAGLLM=Gemma2-27B, Single-step retrieval=true2025.12—88.396.6—
CoopRAGLLM=Llama3.3-70B, Single-step retrieval=true2025.12—86.996.6—
CoopRAGLLM=GPT-4o-mini, Single-step retrieval=true2025.12—88.896.8—
GritLM-7BMethod Type=Large Embedding Model, Source=Literature2026.06—79.292.4—
GTE-Qwen2-7B-InstructMethod Type=Large Embedding Model, Source=Literature2026.06—75.889.1—
GTR (T5-based)Method Type=Simple Baseline, Source=Literature2026.06—59.373.9—
HippoRAG 2Method Type=Graph-based Method, Embedding Model=BGE-Large-EN-v1.5, LLM=Qwen3.6-Flash2026.06—78.494.4—
HippoRAG2LLM=Llama3.3-70B, Single-step retrieval=true2025.12—83.596.3—
HippoRAG2LLM=GPT-4o-mini, Single-step retrieval=true2025.12—80.595.7—
HIPPORAG22026.04——96.31.1
HopRAGLLM=GPT-4o-mini, Single-step retrieval=true2025.12—81.196—
NV-Embed-v2 (7B)Method Type=Large Embedding Model, Source=Literature2026.06—84.194.5—
PROPRAG2026.04——97.4—
SAGMethod Type=Graph-based Method, Embedding Model=BGE-Large-EN-v1.5, LLM=Qwen3.6-Flash2026.06—91.696.5—
SiReRAGLLM=GPT-4o-mini, Single-step retrieval=true2025.12—8094.8—