Information Retrieval on BRIGHT 1.0 (test)
37.9nDCG@10 (Avg)Rank-K-32B*
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Rank-K-32B*Training=R1-distill, Retrieval Source=BM25+GPT-4o CoT Top-100 docs, Listwise reranking=over top-20 documents2025.06 | 37.9 | 50.8 | 49.4 | 28.2 | 46 | 27.3 | 30.5 | 31.9 | — | — | — | — | — | — | — | |
| ThinkQE (Phi-4-Reasoning-14B)Training=Zero-shot2025.06 | 36 | 51.8 | 53.5 | 29.7 | 38.5 | 21.8 | 29.3 | 27.7 | — | — | — | — | — | — | — | |
| ThinkQE (QWEN3-8B)Training=Zero-shot2025.06 | 35.2 | 49.8 | 55.3 | 27.6 | 36.7 | 19.9 | 29 | 28.3 | — | — | — | — | — | — | — | |
| ThinkQE (R1-14B)Training=Zero-shot2025.06 | 34.9 | 47.3 | 52.5 | 29.2 | 40 | 19.3 | 28 | 27.9 | — | — | — | — | — | — | — | |
| ThinkQE (OpenThinker2-7B)Training=Zero-shot2025.06 | 34.6 | 50.5 | 54.1 | 25.8 | 36.7 | 18.1 | 28.2 | 28.9 | — | — | — | — | — | — | — | |
| BM25 + GPT-4o CoTTraining=Zero-shot2025.06 | 33.9 | 53.6 | 53.6 | 24.3 | 38.6 | 18.8 | 22.7 | 25.9 | — | — | — | — | — | — | — | |
| ThinkQE (QWEN3-14B)Training=Zero-shot2025.06 | 33.6 | 51.5 | 53.2 | 27.8 | 37.2 | 22 | 16.1 | 27.5 | — | — | — | — | — | — | — | |
| Rank1-14BTraining=R1-distill, Retrieval Source=BM25+GPT-4o CoT Top-100 docs2025.06 | 31.7 | 49.3 | 37.7 | 22.6 | 35.2 | 22.5 | 20.8 | 33.6 | — | — | — | — | — | — | — | |
| LameR-R1-14BTraining=Zero-shot, Category=Query expansion with BM252025.06 | 29.3 | 35.1 | 46.1 | 23.7 | 31 | 17.7 | 26.4 | 25.3 | — | — | — | — | — | — | — | |
| REPAIR-7BReranking=top-100 ReasonIR results2026.01 | 29 | 44.4 | 36.7 | 25.4 | 41.2 | 26.7 | 18.3 | 28.7 | 23.9 | 27.7 | 7.4 | 34.1 | 33.9 | — | — | |
| REPAIR-7BTraining Context=Training with reasoning-intensive data, Reranking top-100=true2026.01 | 28.3 | 43.1 | 45 | 32.4 | 32.5 | 30.9 | 26.3 | 36.5 | 15.2 | 31.4 | 8.3 | 17.9 | 19.8 | — | — | |
| HyDE-R1-14BTraining=Zero-shot, Category=Query expansion with BM252025.06 | 27.2 | 33.3 | 44.9 | 21.1 | 29.8 | 16.3 | 24.1 | 21 | — | — | — | — | — | — | — | |
| Rank-R1-14BTraining=GRPO (RL), Retrieval Source=BM25 Top-100 docs2025.06 | 26.6 | 31.2 | 38.5 | 21.2 | 26.4 | 22.6 | 18.9 | 27.5 | — | — | — | — | — | — | — | |
| ReasonRank-7BTraining Context=Training with reasoning-intensive data, Reranking top-100=true2026.01 | 26.5 | 36.1 | 44 | 24.7 | 30.4 | 31.1 | 25.3 | 33.1 | 17.7 | 30 | 11.5 | 18.7 | 14.8 | — | — | |
| ReasonIR-8BTraining=SFT2025.06 | 24.8 | 26.2 | 31.4 | 23.3 | 30 | 18 | 23.9 | 20.5 | — | — | — | — | — | — | — | |
| REPAIR-7BTraining Context=Training with knowledge-intensive data, Reranking top-100=true2026.01 | 24.7 | 39.6 | 38.1 | 21.4 | 32.6 | 24.6 | 17.2 | 33.1 | 15.6 | 29 | 7.2 | 16.9 | 21.2 | — | — | |
| RankGPT4Training=Zeroshot, Retrieval Source=BM25 Top-100 docs2025.06 | 24.7 | 33.8 | 34.2 | 16.7 | 27 | 22.3 | 27.7 | 11.1 | — | — | — | — | — | — | — | |
| REARANK-7BReranking=top-100 ReasonIR results2026.01 | 24.6 | 28.8 | 25.4 | 24.7 | 31.5 | 19.4 | 18.8 | 28.5 | 13.1 | 33.8 | 10.5 | 29.7 | 30.5 | — | — | |
| ReasonIRInitial retrieval=true2026.01 | 24.3 | 26.3 | 31.5 | 23.3 | 30.3 | 18 | 23.5 | 20.6 | 10.3 | 35 | 14.3 | 31.6 | 27.2 | — | — | |
| QwenRerank-7BReranking=top-100 ReasonIR results2026.01 | 24.2 | 31.9 | 40.7 | 20.3 | 31.2 | 18 | 23.4 | 24.6 | 18.6 | 27.1 | 6.5 | 23.9 | 33.1 | — | — | |
| Rank-R1-7BReranking=top-100 ReasonIR results2026.01 | 24.2 | 39.5 | 25.7 | 22.8 | 33.4 | 18.6 | 19.6 | 29.6 | 9.3 | 21.4 | 7.6 | 26.7 | 36.6 | — | — | |
| RankZephyr-7BReranking=top-100 ReasonIR results2026.01 | 23.3 | 25.8 | 31.5 | 22.3 | 30 | 16.4 | 21 | 20.6 | 10.1 | 33 | 13.1 | 29.9 | 26.4 | — | — | |
| GTE-QWEN-7BTraining=SFT2025.06 | 22.8 | 30.6 | 36.4 | 17.8 | 24.6 | 13.2 | 22.2 | 14.8 | — | — | — | — | — | — | — | |
| QwenRerank-7BTraining Context=Zero-shot, Initial Retrieval=false, Reranking top-100=true2026.01 | 21.2 | 28 | 40.8 | 16.7 | 26.4 | 24.8 | 24 | 24.9 | 12.1 | 24.8 | 6.7 | 12.3 | 13.3 | — | — | |
| GritLM-7BTraining=SFT2025.06 | 20.6 | 24.8 | 32.3 | 18.9 | 19.8 | 17.1 | 13.6 | 17.8 | — | — | — | — | — | — | — | |
| PSRCategory=Planning-based Reranking2026.01 | 20.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| REARANK-7BTraining Context=Training with knowledge-intensive data, Reranking top-100=true2026.01 | 19.1 | 22.6 | 25.4 | 15.9 | 21.4 | 22.6 | 17.5 | 27 | 14 | 27.8 | 6.2 | 16.6 | 12.6 | — | — | |
| Rank-R1-7BTraining Context=Training with knowledge-intensive data, Reranking top-100=true2026.01 | 17.9 | 29.2 | 29.8 | 17.2 | 24.2 | 22.8 | 11.2 | 26.6 | 8.3 | 16.3 | 3.2 | 12.2 | 13.5 | — | — | |
| BM25Training=Zero-shot2025.06 | 17 | 18.2 | 27.9 | 16.4 | 13.4 | 10.9 | 16.3 | 16.1 | — | — | — | — | — | — | — | |
| RankZephyr-7bTraining=GPT4-distill, Retrieval Source=BM25 Top-100 docs2025.06 | 15.5 | 21.9 | 23.7 | 14.4 | 10.3 | 7.6 | 13.7 | 16.6 | — | — | — | — | — | — | — | |
| RL-IndexRetriever=BGE, Augmentation Method=RL-Index, Document rationale augmentor=Llama-3.2-3B-Instruct2026.06 | 15.4 | 14.1 | 27.2 | 16.9 | 18.9 | 14 | 14 | 14 | 10.5 | — | 5.9 | 9.6 | 13.9 | 26 | 13.2 | |
| BM25Training Context=Zero-shot, Initial Retrieval=true2026.01 | 14.5 | 18.7 | 27.2 | 14.9 | 12.5 | 13.6 | 18.4 | 15 | 7.9 | 24.4 | 6.2 | 10.4 | 4.9 | — | — | |
| SPIKERetriever=BGE, Augmentation Method=SPIKE2026.06 | 14.4 | 13.2 | 26.4 | 17 | 18.1 | 11.5 | 13.3 | 13.2 | 6.4 | — | 4.8 | 8.5 | 13 | 27.1 | 5.9 | |
| SPIKE*Retriever=BGE, Augmentation Method=SPIKE*2026.06 | 14 | 13 | 24.4 | 13.3 | 18 | 12.2 | 13.1 | 13.5 | 7.7 | — | 5.5 | 8 | 12.7 | 26 | 3 | |
| BGERetriever=BGE2026.06 | 13.6 | 11.7 | 24.4 | 16.4 | 17.4 | 11.7 | 10.6 | 13.1 | 5.7 | — | 6 | 6.9 | 13 | 26.7 | — | |
| RankZephyr-7BTraining Context=Training with knowledge-intensive data, Reranking top-100=true2026.01 | 13.1 | 19.2 | 21.9 | 13.9 | 10.8 | 13.9 | 14.9 | 15 | 6.4 | 22.6 | 4.4 | 9.1 | 4.6 | — | — | |
| Doc2QueryRetriever=BGE, Augmentation Method=Doc2Query, Number of synthetic queries (k)=102026.06 | 13 | 8.7 | 24.2 | 15.3 | 16.9 | 13.6 | 10.9 | 12.7 | 3.7 | — | 6.7 | 6 | 11.7 | 25.4 | -4.4 | |
| Doc2QueryRetriever=BGE, Augmentation Method=Doc2Query, Number of synthetic queries (k)=32026.06 | 12.8 | 9.5 | 23.9 | 15.5 | 17.3 | 10.6 | 10.5 | 13.1 | 3.8 | — | 7 | 4.5 | 12.6 | 25.6 | -5.9 | |
| SUNAR_RCategory=Conventional NAR, Augmented with SAR plan=true2026.01 | 11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SUNAR_RReranking=top-100 ReasonIR results2026.01 | 10.7 | 7.8 | 25.2 | 6.6 | 7.4 | 3.7 | 7.1 | 8 | 32 | 12.3 | 5.8 | 5.8 | 6.7 | — | — | |
| SUNAR_RTraining Context=Zero-shot, Initial Retrieval=false, Reranking top-100=true2026.01 | 10.1 | 16.9 | 23.2 | 9.4 | 10.5 | 10 | 11.2 | 12.9 | 6.3 | 9.5 | 2.7 | 5.4 | 3.6 | — | — |