Elamur Research⌕
☰
home / papers / 2025 / QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management

QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management

Paper · 2025 · Weizhou Shen, Ziyi Yang, Chenliang Li et al.

Summary

Research on long contexts has mostly targeted pre-training and architecture, so post-training lags behind. Hard long-document training data is scarce, RL becomes unstable on long multi-task sequences, and some inputs exceed any context window. The Tongyi Lab (Alibaba) team presents an end-to-end recipe applied to Qwen3-30B-A3B-Thinking.

First, a synthesis pipeline splits documents into atomic facts and the links between them. It then programmatically builds verifiable multi-hop questions whose evidence is spread across the whole text, which goes well beyond needle-style retrieval.

Second, GRPO-based RL is stabilized in three ways: balanced sampling across tasks, advantages estimated separately per task, and a new algorithm, AEPO, which uses policy entropy to regulate negative gradients as training sequences grow longer.

Third, a memory agent reads documents chunk by chunk, updating a memory and a plan at each step. Multi-stage fusion RL with model merging combines this with native 256K-window reasoning.

The resulting model rivals GPT-5 and Gemini-2.5-Pro and also improves on math, tool use and long dialogue. It offers open models a practical route to strong long-context reasoning.

Key points

  • QwenLong-L1.5-30B-A3B gains 9.9 points on average over Qwen3-30B-A3B-Thinking across six benchmarks: MRCR, CorpusQA, DocMath, LongBench-V1-QA, Frames and LongBench-V2. It outperforms DeepSeek-R1-0528, Gemini-2.5-Flash-Thinking and Qwen3-Max-Thinking.
  • On 1M–4M-token tasks, the memory-agent mode beats the baseline's agent setup by 9.48 points and its single-pass setup by 15.26 points.
  • The RL training set grows from 1.6K samples (QwenLong-L1, open-source only) to 14.1K samples mixing open-source and synthetic data.
  • RL setup: GRPO with DAPO's token-level loss, no KL penalty (β = 0), strictly on-policy single updates, and AEPO for entropy control.
  • On MRCR-1M, the memory-RL stage lowered full-context accuracy to 68.53. Merging restored it to 71.18 and lifted the memory-agent score to 21.68, and a final full-context RL stage raised that score to 22.53.
  • Code: github.com/Tongyi-Zhiwen/Qwen-Doc.

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

🏢Alibaba GroupHangzhou · China168 works · 15,386 citations🏢Alibaba Group (Cayman Islands)George Town · Cayman Islands32 works · 3,389 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 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Weizhou Shen, Ziyi Yang, Chenliang Li, Zhiyuan Lu, Miao Peng, Huashan Sun, Yingcheng Shi, Shengyi Liao, Shaopeng Lai, Bo Zhang, Dayiheng Liu, Fei Huang, Jingren Zhou and Ming Yan. 2025. QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management. arXiv:2512.12967.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2512.12967 · arXiv 2512.12967

@misc{shen2025qwenlong,
  title = {{QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management}},
  author = {Weizhou Shen, Ziyi Yang, Chenliang Li, Zhiyuan Lu, Miao Peng, Huashan Sun, Yingcheng Shi, Shengyi Liao, Shaopeng Lai, Bo Zhang, Dayiheng Liu, Fei Huang, Jingren Zhou and Ming Yan},
  year = {2025},
  doi = {10.48550/arxiv.2512.12967},
  eprint = {2512.12967},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2512.12967}
}

References

50 references, 27 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dilip Arumugam, Peter Henderson, and Pierre-Luc Bacon. An information-theoretic perspective on credit assignment in reinforcement learning. ArXiv, abs/2103.06224, 2021. URL https://api.semanticscholar.org/CorpusID:228215123. 10 — [arXiv:2103.06224](https://arxiv.org/abs/2103.06224) ↗
  2. Yushi Bai, Xin Lv, Jiajie Zhang, Hongchang Lyu, Jiankai Tang, Zhidian Huang, Zhengxiao Du, Xiao Liu, Aohan Zeng, Lei Hou, Yuxiao Dong, Jie Tang, and Juanzi Li. LongBench: A bilingual, multitask benchmark for long context understanding. In Proceedings of the 62nd Annual Meeting of the Association for…
  3. Yushi Bai, Shangqing Tu, Jiajie Zhang, Hao Peng, Xiaozhi Wang, Xin Lv, Shulin Cao, Jiazheng Xu, Lei Hou, Yuxiao Dong, Jie Tang, and Juanzi Li. LongBench v2: Towards deeper understanding and reasoning on realistic long-context multitasks. In Proceedings of the 63rd Annual Meeting of the Association f…
  4. Elie Bakouch, Leandro von Werra, and Lewis Tunstall. Open-r1: a fully open reproduction of deepseek-r1. URL https://huggingface.co/blog/open-r1. 9 Victor Barres, Honghua Dong, Soham Ray, Xujie Si, and Karthik Narasimhan. τ 2 -bench: Evaluating conversational agents in a dual-control environment, 202… ↗
  5. Pradeep Dasigi, Kyle Lo, Iz Beltagy, Arman Cohan, Noah A Smith, and Matt Gardner. A dataset of information-seeking questions and answers anchored in research papers. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language T…
  6. Deepmind. Gemini 2.5: Our most intelligent ai model. Technical report, Deepmind, 2025. URL https://blog.google/technology/google-deepmind/ gemini-model-thinking-updates-march-2025/. 2 ↗
  7. Xanh Ho, Anh-Khoa Duong Nguyen, Saku Sugawara, and Akiko Aizawa. Constructing a multi-hop qa dataset for comprehensive evaluation of reasoning steps. In Proceedings of the 28th International Conference on Computational Linguistics, 2020. 12
  8. Cheng-Ping Hsieh, Simeng Sun, Samuel Kriman, Shantanu Acharya, Dima Rekesh, Fei Jia, Yang Zhang, and Boris Ginsburg. Ruler: What’s the real context size of your long-context language models?, 2024. URL https://arxiv.org/abs/2404.06654. 2 ↗
  9. Jian Hu. Reinforce++: A simple and efficient approach for aligning large language models. arXiv preprint arXiv:2501.03262, 2025. 8 find
  10. Jingcheng Hu, Yinmin Zhang, Qi Han, Daxin Jiang, Xiangyu Zhang, and Heung-Yeung Shum. Open-reasoner-zero: An open source approach to scaling up reinforcement learning on the base model. arXiv preprint arXiv:2503.24290, 2025. 3
  11. Gregory Kamradt. Needle in a haystack - pressure testing llms. https://github.com/gkamradt/ LLMTestNeedleInAHaystack/tree/main, 2023. 2 ↗
  12. Tomáš Kočiskỳ, Jonathan Schwarz, Phil Blunsom, Chris Dyer, Karl Moritz Hermann, Gábor Melis, and Edward Grefenstette. The narrativeqa reading comprehension challenge. Transactions of the Association for Computational Linguistics, 6:317–328, 2018. 2, 12
  13. Satyapriya Krishna, Kalpesh Krishna, Anhad Mohananey, Steven Schwarcz, Adam Stambler, Shyam Upadhyay, and Manaal Faruqui. Fact, fetch, and reason: A unified evaluation of retrieval-augmented generation. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association f… find
  14. Jiawei Li, Yang Gao, Yizhe Yang, Yu Bai, Xiaofeng Zhou, Yinghao Li, Huashan Sun, Yuhang Liu, Xingpeng Si, Yuhao Ye, Yixiao Wu, Yiguan Lin, Bin Xu, Bowen Ren, Chong Feng, and Heyan Huang. Fundamental capabilities and applications of large language models: A survey. ACM Comput. Surv., 58(2), September… — [doi:10.1145/3735632](https://doi.org/10.1145/3735632) find
  15. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437, 2024. 13 find
  16. Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, and Min Lin. Understanding r1-zero-like training: A critical perspective. In Conference on Language Modeling (COLM), 2025a. 8
  17. Zihe Liu, Jiashun Liu, Yancheng He, Weixun Wang, Jiaheng Liu, Ling Pan, Xinyu Hu, Shaopan Xiong, Ju Huang, Jian Hu, et al. Part i: Tricks or traps? a deep dive into rl for llm reasoning. arXiv preprint arXiv:2508.08221, 2025b. 8 find
  18. MAA. American invitational mathematics examination - AIME, 2025. maa-invitational-competitions/. 14 find
  19. URL — UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction https://maa.org/Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426, 2018. 9 ↗
  20. Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. Gaia: a benchmark for general ai assistants, 2023. URL https://arxiv.org/abs/2311.12983. 2 ↗
  21. OpenAI. gpt-oss-120b & gpt-oss-20b model card, 2025. URL https://cdn.openai.com/pdf/ 419b6906-9da6-406c-a19d-1bb078ac7637/oai_gpt-oss_model_card.pdf. 12 ↗
  22. Shishir G. Patil, Huanzhi Mao, Charlie Cheng-Jie Ji, Fanjia Yan, Vishnu Suresh, Ion Stoica, and Joseph E. Gonzalez. The berkeley function calling leaderboard (bfcl): From tool use to agentic evaluation of large language models. In Forty-second International Conference on Machine Learning, 2025. 2, 1… find
  23. Zihan Qiu, Zekun Wang, Bo Zheng, Zeyu Huang, Kaiyue Wen, Songlin Yang, Rui Men, Le Yu, Fei Huang, Suozhi Huang, Dayiheng Liu, Jingren Zhou, and Junyang Lin. Gated attention for large language models: Non-linearity, sparsity, and attention-sink-free, 2025. URL https://arxiv.org/abs/2505.06708. 2 — [arXiv:2505.06708](https://arxiv.org/abs/2505.06708) ↗
  24. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R. Bowman. Gpqa: A graduate-level google-proof q&a benchmark, 2023. URL https://arxiv.org/abs/2311.12022. 14 ↗
  25. John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel. High-dimensional continuous control using generalized advantage estimation. arXiv preprint arXiv:1506.02438, 2015. 3 find
  26. John Schulman, Xi Chen, and Pieter Abbeel. Equivalence between policy gradients and soft q-learning. arXiv preprint arXiv:1704.06440, 2017a. 2 find
  27. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017b. 3 find
  28. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024. 3
  29. Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. Hybridflow: A flexible and efficient rlhf framework. In Proceedings of the Twentieth European Conference on Computer Systems, 2025. 11
  30. Huatong Song, Jinhao Jiang, Wenqing Tian, Zhipeng Chen, Yuhuan Wu, Jiahao Zhao, Yingqian Min, Wayne Xin Zhao, Lei Fang, and Ji-Rong Wen. R1-searcher++: Incentivizing the dynamic knowledge acquisition of llms via reinforcement learning. arXiv preprint arXiv:2505.17005, 2025. 12 — [arXiv:2505.17005](https://arxiv.org/abs/2505.17005) find
  31. Huashan Sun, Shengyi Liao, Yansen Han, Yu Bai, Yang Gao, Cheng Fu, Weizhou Shen, Fanqi Wan, Ming Yan, Ji Zhang, et al. Solopo: Unlocking long-context capabilities in llms via short-to-long preference optimization. arXiv preprint arXiv:2505.11166, 2025. 10 — [arXiv:2505.11166](https://arxiv.org/abs/2505.11166) find
  32. Richard S. Sutton and Andrew G. Barto. Reinforcement learning: An introduction. IEEE Trans. Neural Networks, 9:1054–1054, 1998. URL https://api.semanticscholar.org/CorpusID:60035920. 10 ↗
  33. Kimi Team, Yifan Bai, Yiping Bao, Guanduo Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, et al. Kimi k2: Open agentic intelligence. arXiv preprint arXiv:2507.20534, 2025a. 2
  34. Kimi Team, Yu Zhang, Zongyu Lin, Xingcheng Yao, Jiaxi Hu, Fanqing Meng, Chengyin Liu, Xin Men, Songlin Yang, Zhiyuan Li, Wentao Li, Enzhe Lu, Weizhou Liu, Yanru Chen, Weixin Xu, Longhui Yu, Yejie Wang, Yu Fan, Longguang Zhong, Enming Yuan, Dehao Zhang, Yizhi Zhang, T. Y. Liu, Haiming Wang, Shengjun… — Kimi Linear: An Expressive, Efficient Attention Architecture
  35. Tongyi-Zhiwen Team. Corpusqa: A 10 million token benchmark for corpus-level analysis and reasoning. https://github.com/Tongyi-Zhiwen/CorpusQA, 2025. 6, 12 ↗
  36. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. Musique: Multihop questions via single-hop question composition. Transactions of the Association for Computational Linguistics, 10:539–554, 2022. 12 — ♫ MuSiQue: Multihop Questions via Single-hop Question Composition
  37. Kiran Vodrahalli, Santiago Ontanon, Nilesh Tripuraneni, Kelvin Xu, Sanil Jain, Rakesh Shivanna, Jeffrey Hui, Nishanth Dikkala, Mehran Kazemi, Bahare Fatemi, et al. Michelangelo: Long context evaluations beyond haystacks via latent structure queries. arXiv preprint arXiv:2409.12640, 2024. 2, 12
  38. Fanqi Wan, Longguang Zhong, Ziyi Yang, Ruijun Chen, and Xiaojun Quan. Fusechat: Knowledge fusion of chat models, 2024. URL https://arxiv.org/abs/2408.07990. 8 ↗
  39. Fanqi Wan, Weizhou Shen, Shengyi Liao, Yingcheng Shi, Chenliang Li, Ziyi Yang, Ji Zhang, Fei Huang, Jingren Zhou, and Ming Yan. Qwenlong-l1: Towards long-context large reasoning models with reinforcement learning. arXiv preprint arXiv:2505.17667, 2025. 2, 3, 8, 12, 13
  40. Jiawei Wang, Jiacai Liu, Yuqian Fu, Yingru Li, Xintao Wang, Yuan Lin, Yu Yue, Lin Zhang, Yang Wang, and Ke Wang. Harnessing uncertainty: Entropy-modulated policy gradients for long-horizon llm agents, 2025a. URL https://arxiv.org/abs/2509.09265. 10 — [arXiv:2509.09265](https://arxiv.org/abs/2509.09265) ↗
  41. Shenzhi Wang, Le Yu, Chang Gao, Chujie Zheng, Shixuan Liu, Rui Lu, Kai Dang, Xionghui Chen, Jianxin Yang, Zhenru Zhang, et al. Beyond the 80/20 rule: High-entropy minority tokens drive effective reinforcement learning for llm reasoning. arXiv preprint arXiv:2506.01939, 2025b. 10
  42. Yubo Wang, Xueguang Ma, Ge Zhang, Yuansheng Ni, Abhranil Chandra, Shiguang Guo, Weiming Ren, Aaran Arulraj, Xuan He, Ziyan Jiang, Tianle Li, Max Ku, Kai Wang, Alex Zhuang, Rongqi Fan, Xiang Yue, and Wenhu Chen. MMLU-pro: A more robust and challenging multi-task language understanding benchmark. In T…
  43. Di Wu, Hongwei Wang, Wenhao Yu, Yuwei Zhang, Kai-Wei Chang, and Dong Yu. Longmemeval: Benchmarking chat assistants on long-term interactive memory. 2024. URL https://arxiv.org/abs/2410.10813. 2, 14 ↗
  44. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report. arXiv preprint arXiv:2505.09388, 2025a. 2, 11
  45. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D Manning. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018. 12
  46. Ziyi Yang, Weizhou Shen, Ruijun Chen, Chenliang Li, Fanqi Wan, Ming Yan, Xiaojun Quan, and Fei Huang. Spell: Self-play reinforcement learning for evolving long-context language models. arXiv preprint arXiv:2509.23863, 2025b. 7 — [arXiv:2509.23863](https://arxiv.org/abs/2509.23863) find
  47. Hongli Yu, Tinghong Chen, Jiangtao Feng, Jiangjie Chen, Weinan Dai, Qiying Yu, Ya-Qin Zhang, Wei-Ying Ma, Jingjing Liu, Mingxuan Wang, et al. Memagent: Reshaping long-context llm with multi-conv rl-based memory agent. arXiv preprint arXiv:2507.02259, 2025a. 3, 15 — [arXiv:2507.02259](https://arxiv.org/abs/2507.02259) find
  48. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476, 2025b. 3, 9
  49. Yilun Zhao, Yitao Long, Hongjun Liu, Ryo Kamoi, Linyong Nan, Lyuhao Chen, Yixin Liu, Xiangru Tang, Rui Zhang, and Arman Cohan. Docmath-eval: Evaluating math reasoning capabilities of llms in understanding long and specialized documents. In Proceedings of the 62nd Annual Meeting of the Association fo… find
  50. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems, 2023. 13

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shen-2025-qwenlong-l1-5-post/