Elamur Research⌕
☰
home / papers / 2026 / GroupRank: A Groupwise Paradigm for Effective and Efficient Passage Reranking with LLMs

GroupRank: A Groupwise Paradigm for Effective and Efficient Passage Reranking with LLMs

Paper · 2026 · Meixiu Long, Duolin Sun, Dan Yang et al.

Summary

LLM rerankers face a trade-off between accuracy and speed. Pointwise scoring is fast but never compares documents with each other. Listwise ranking does compare them, but it exceeds context limits and relies on slow sequential sliding windows. GroupRank takes a middle path: it splits the candidate pool into small disjoint groups (about 10 documents each), scores each group jointly and in parallel, and then merges all scores into one ranking.

Training starts with a synthetic dataset of about 14k samples, built without gold answers. Candidates come from hybrid BM25 + dense retrieval, and their pseudo-labels blend pointwise relevance scores with listwise orderings. The model first goes through SFT to learn the output format. It then receives GRPO-based RL with a Group-Ranking reward, which combines a ranking-utility term (rewarding correct ordering) and a group-alignment term (calibrating the score distribution).

GroupRank sets new state-of-the-art results on the reasoning-heavy BRIGHT and R2MED benchmarks while running several times faster than pointwise and listwise baselines. This makes it practical for latency-sensitive retrieval pipelines.

Key points

  • Headline results: 65.2 NDCG@10 on BRIGHT and 52.3 on R2MED, a margin of up to 2.1 points over baselines. The BRIGHT score uses INF-X-Retriever, rewritten queries, and test-time scaling, with gains levelling off after about four iterations.
  • Speed: inference is 2.4–6.4× faster than the baselines on 32B models running on four H800 GPUs.
    • Overall, GroupRank needs 3.4 s/query versus 11.1 s for ERank.
    • On theorem-based tasks, it needs 6.1 s versus 28.7 s for ReasonRank.
  • Training queries: about 13,000 general reasoning queries, 1,000 in-house medical queries, and 700 xCodeEval coding queries. Candidates are the top-50 documents from BM25 + dense fusion with α = 0.5.
  • Group formation: random grouping beats grouping by initial rank, both on BRIGHT (38.0 vs 36.2 avg) and on R2MED (52.3 vs 49.8).
  • Limitations: random grouping can place several highly relevant documents in the same group, which biases scores in a single inference pass. The reasoning-based backbone (Qwen2.5-32B) also adds latency, so smaller non-reasoning models are left for future work.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Sun Yat-sen UniversityGuangzhou · China53 works · 9,185 citations🏢Ant GroupHangzhou · China42 works · 1,131 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksInformation Systems · 1,437 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 473 works

Cite

Cite as Meixiu Long, Duolin Sun, Dan Yang, Yihan Jiao, Lei Liu, Jiahai Wang, Binbin Hu, Yue Shen, Jie Feng, Zhehao Tan, Junjie Wang, Lianzhen Zhong, Jian Wang, Peng Wei and Jinjie Gu. 2026. GroupRank: A Groupwise Paradigm for Effective and Efficient Passage Reranking with LLMs. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1260

@inproceedings{long2026grouprank,
  title = {{GroupRank: A Groupwise Paradigm for Effective and Efficient Passage Reranking with LLMs}},
  author = {Meixiu Long, Duolin Sun, Dan Yang, Yihan Jiao, Lei Liu, Jiahai Wang, Binbin Hu, Yue Shen, Jie Feng, Zhehao Tan, Junjie Wang, Lianzhen Zhong, Jian Wang, Peng Wei and Jinjie Gu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1260},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1260}
}

References

28 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Yuzheng Cai, Yanzhao Zhang, Dingkun Long, Mingxin Li, Pengjun Xie, and Weiguo Zheng. 2025. Erank: Fusing supervised fine-tuning and reinforcement learning for effective and efficient text reranking. CoRR, abs/2509.00520. — [arXiv:2509.00520](https://arxiv.org/abs/2509.00520) find
  2. Mingyue Cheng, Yucong Luo, Jie Ouyang, Qi Liu, Huijie Liu, Li Li, Shuo Yu, Bohou Zhang, Jiawei Cao, Jie Ma, Daoyu Wang, and Enhong Chen. 2025. A survey on knowledge-oriented retrieval-augmented generation. Preprint, arXiv:2503.10677. — [arXiv:2503.10677](https://arxiv.org/abs/2503.10677) find
  3. DeepSeek-AI. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. CoRR, abs/2501.12948.
  4. Chunjing Gan, Dan Yang, Binbin Hu, Hanxiao Zhang, Siyuan Li, Ziqi Liu, Yue Shen, Lin Ju, Zhiqiang Zhang, Jinjie Gu, Lei Liang, and Jun Zhou. 2024. Similarity is not all you need: Endowing retrieval augmented generation with multi layered thoughts. find
  5. Mohammad Abdullah Matin Khan, M. Saiful Bari, Xuan Do Long, Weishi Wang, Md. Rizwan Parvez, and Shafiq Joty. 2024. Xcodeeval: An execution-based large scale multilingual multitask benchmark for code understanding, generation, translation and retrieval. In ACL (1), pages 6766–6805.
  6. S. Kullback and R. A. Leibler. 1951. On information and sufficiency. The Annals of Mathematical Statistics, 22(1):79–86. find
  7. Junwei Lan, Jianlyu Chen, Zheng Liu, Chaofan Li, Siqi Bao, and Defu Lian. 2025. Retro*: Optimizing LLMs for reasoning-intensive document retrieval. CoRR, abs/2509.24869. — [arXiv:2509.24869](https://arxiv.org/abs/2509.24869) find
  8. Lei Li, Xiao Zhou, and Zheng Liu. 2025a. R2med: A benchmark for reasoning-driven medical retrieval. Preprint, arXiv:2505.14558. — [arXiv:2505.14558](https://arxiv.org/abs/2505.14558) find
  9. Yuchen Li, Hao Zhang, Yongqi Zhang, Xinyu Ma, Wenwen Ye, Naifei Song, Shuaiqiang Wang, Haoyi Xiong, Dawei Yin, and Lei Chen. 2025b. M2 oerank: Multi-objective mixture-of-experts enhanced ranking for satisfaction-oriented web search. In ICDE, pages 4441–4454. find
  10. Jimmy Lin, Xueguang Ma, Sheng-Chieh Lin, Jheng-Hong Yang, Ronak Pradeep, and Rodrigo Nogueira. 2021. Pyserini: A python toolkit for reproducible information retrieval research with sparse and dense representations. In SIGIR, pages 2356–2362. 25173 9 find
  11. Wenhan Liu, Xinyu Ma, Weiwei Sun, Yutao Zhu, Yuchen Li, Dawei Yin, and Zhicheng Dou. 2025a. Reasonrank: Empowering passage ranking with strong reasoning ability. CoRR, abs/2508.07050.
  12. Orion Weller, Kathryn Ricci, Eugene Yang, Andrew Yates, Dawn J. Lawrie, and Benjamin Van Durme. 2025. Rank1: Test-time compute for reranking in information retrieval. CoRR, abs/2502.18418. — [arXiv:2502.18418](https://arxiv.org/abs/2502.18418) find
  13. Wenhan Liu, Xinyu Ma, Yutao Zhu, Ziliang Zhao, Shuaiqiang Wang, Dawei Yin, and Zhicheng Dou. 2025b. Sliding windows are not the end: Exploring full ranking with long-context large language models. In ACL (1), pages 162–176. find
  14. Juan Wisznia, Cecilia Bolaños, Juan Tollo, Giovanni Marraffini, Agustín Gianolini, Noe Hsueh, and Luciano Del Corro. 2025. Are optimal algorithms still optimal? rethinking sorting in llm-based pairwise ranking with batching and caching. In ACL (2), pages 1064–1072. Association for Computational Ling… find
  15. Meixiu Long, DuoLin Sun, Dan Yang, Junjie Wang, Yue Shen, Jian Wang, Peng Wei, Jinjie Gu, and Jiahai Wang. 2025. DIVER: A multi-stage approach for reasoning-intensive information retrieval. CoRR, abs/2508.07995. — [arXiv:2508.07995](https://arxiv.org/abs/2508.07995) find
  16. Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. 2016. MS MARCO: A human generated machine reading comprehension dataset. In CoCo@NIPS, volume 1773 of CEUR Workshop Proceedings.
  17. Tong Niu, Shafiq Joty, Ye Liu, Caiming Xiong, Yingbo Zhou, and Semih Yavuz. 2024. Judgerank: Leveraging large language models for reasoning-intensive reranking. CoRR, abs/2411.00142. — [arXiv:2411.00142](https://arxiv.org/abs/2411.00142) find
  18. Ronak Pradeep, Sahel Sharifymoghaddam, and Jimmy Lin. 2023. Rankzephyr: Effective and robust zero-shot listwise reranking is a breeze! CoRR, abs/2312.02724. — [arXiv:2312.02724](https://arxiv.org/abs/2312.02724) find
  19. Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, and 25 others. 2025. Qwen2.5 technical report. Preprint, arXiv:2412.15115. find
  20. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, and 40 others. 2025a. Qwen3 technical report. CoRR, abs/2505.09388.
  21. Eugene Yang, Andrew Yates, Kathryn Ricci, Orion Weller, Vivek Chari, Benjamin Van Durme, and Dawn J. Lawrie. 2025b. Rank-k: Test-time reasoning for listwise reranking. CoRR, abs/2505.14432. — [arXiv:2505.14432](https://arxiv.org/abs/2505.14432) find
  22. Yichen Yao, Jiahe Wan, Yuxin Hong, Mengna Zhang, Junhan Yang, Zhouyu Jiang, Qing Xu, Kuan Lu, Yinghui Xu, Wei Chu, and Yuan Qi. 2025. Inf-x-retriever. Online. GitHub repository. find
  23. Honglei Zhuang, Zhen Qin, Rolf Jagerman, Kai Hui, Ji Ma, Jing Lu, Jianmo Ni, Xuanhui Wang, and Michael Bendersky. 2023. Rankt5: Fine-tuning T5 for text ranking with ranking losses. In SIGIR, pages 2308–2313. ACM. find
  24. Shengyao Zhuang, Xueguang Ma, Bevan Koopman, Jimmy Lin, and Guido Zuccon. 2025. Rank-r1: Enhancing reasoning in llm-based document rerankers via reinforcement learning. CoRR, abs/2503.06034. — [arXiv:2503.06034](https://arxiv.org/abs/2503.06034) find
  25. Revanth Gangi Reddy, JaeHyeok Doo, Yifei Xu, Md. Arafat Sultan, Deevya Swain, Avirup Sil, and Heng Ji. 2024. FIRST: faster improved listwise reranking with single token decoding. In EMNLP, pages 8642–8652. Association for Computational Linguistics. find
  26. Hongjin Su, Howard Yen, Mengzhou Xia, Weijia Shi, Niklas Muennighoff, Han-yu Wang, Haisu Liu, Quan Shi, Zachary S. Siegel, Michael Tang, Ruoxi Sun, Jinsung Yoon, Sercan Ö. Arik, Danqi Chen, and Tao Yu. 2025. BRIGHT: A realistic and challenging benchmark for reasoning-intensive retrieval. In ICLR.
  27. Gemini Team. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. CoRR, abs/2507.06261.
  28. Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. Beir: A heterogenous benchmark for zero-shot evaluation of information retrieval models. Preprint, arXiv:2104.08663. 25174 10

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/long-2026-grouprank-groupwise-paradigm-effective-efficient/