Elamur Research⌕
☰
home / papers / 2026 / Self-Sum: Teaching an Agent to Decide Itself When and What to Summarize

Self-Sum: Teaching an Agent to Decide Itself When and What to Summarize

Paper · 2026 · Hongru Wang, Rui Wang, Jushi Kai et al.

Summary

Agents that tackle long tasks pile up reasoning steps, actions and observations. Their context grows noisy and expensive to process. Common fixes compress history on a fixed schedule or hand the job to an external model. These approaches ignore where the agent actually is in the task and can discard crucial details. Self-Sum treats summarizing as an internal action the agent may choose, sitting alongside ordinary environment actions within a multi-turn dialogue between user and agent. Training proceeds in two stages:

  • Supervised warm-up. GPT-5-mini reviews successful trajectories as a whole and marks where a summary fits. It writes grounded summaries, justifies each chosen moment and adds a rationale for every action. The trajectories are then split into conversation segments.
  • Reinforcement learning. A light RL phase tunes when and what to summarize and penalizes superfluous summaries.

On ALFWorld, SciWorld and WebShop, Self-Sum beats both no-summary and rule-based baselines, especially on unseen tasks. Learned, sparse self-summarization therefore looks like a practical route to more robust long-horizon agents.

Key points

  • Baselines span three families: no summarization (e.g., ReAct-style), rule-based compression (ReSum, Verl-Agent) and RL agents such as GiGPO. All experiments use Qwen2.5-series backbones.
  • Generalization is tested at three levels on ALFWorld and SciWorld: L0 (seen), L1 (new variants, known categories) and L2 (new variants and categories). The gain reaches +2.5 absolute points.
  • On WebShop, Self-Sum with Qwen-2.5-1.5B-Instruct scores 89.2 ±0.7 with a 77.1 ±0.5 success rate. This exceeds 7B systems such as GiGPO-7B (86.2 / 75.2) and AgentEvol.
  • In the error analysis, few failures come from bad summaries. Most stem from repeated or invalid actions, which the authors flag as a target for finer credit assignment.
  • Code is released at github.com/BayesWatch/SelfSum.

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

🎓University of EdinburghEdinburgh · United Kingdom359 works · 77,352 citations🎓Tsinghua UniversityBeijing · China333 works · 87,019 citations🎓Hong Kong University of Science and TechnologyHong Kong · China142 works · 25,136 citations🎓Shanghai Jiao Tong UniversityShanghai · China149 works · 16,161 citations🎓Chinese University of Hong KongHong Kong · China126 works · 14,073 citations🎓Hong Kong Polytechnic UniversityHong Kong · China84 works · 4,956 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

🧭 Topics 1Research subjects and languages the work is about.Summarization · 232 works

Cite

Cite as Hongru Wang, Rui Wang, Jushi Kai, Boyang Xue, Yongqi Li, Shijue Huang, Xiaoteng Ma, Jeff Z. Pan and Amos Storkey. 2026. Self-Sum: Teaching an Agent to Decide Itself When and What to Summarize. 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.447

@inproceedings{wang2026self,
  title = {{Self-Sum: Teaching an Agent to Decide Itself When and What to Summarize}},
  author = {Hongru Wang, Rui Wang, Jushi Kai, Boyang Xue, Yongqi Li, Shijue Huang, Xiaoteng Ma, Jeff Z. Pan and Amos Storkey},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.447},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.447}
}

References

35 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Arash Ahmadian, Chris Cremer, Matthias Gallé, Marzieh Fadaee, Julia Kreutzer, Olivier Pietquin, Ahmet Üstün, and Sara Hooker. 2024. Back to basics: Revisiting REINFORCE-style optimization for learning from human feedback in LLMs. In Proceedings of the 62nd Annual Meeting of the Association for Compu… find
  2. Tianzhe Chu, Yuexiang Zhai, Jihan Yang, Shengbang Tong, Saining Xie, Dale Schuurmans, Quoc V Le, Sergey Levine, and Yi Ma. 2025. SFT memorizes, RL generalizes: A comparative study of foundation model post-training. In Forty-second International Conference on Machine Learning.
  3. Tianqing Fang, Hongming Zhang, Zhisong Zhang, Kaixin Ma, Wenhao Yu, Haitao Mi, and Dong Yu. 2025a. Webevolver: Enhancing web agent self-improvement with co-evolving world model. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 8970– 8986.
  4. Tianqing Fang, Zhisong Zhang, Xiaoyang Wang, Rui Wang, Can Qin, Yuxuan Wan, Jun-Yu Ma, Ce Zhang, Jiaqi Chen, Xiyun Li, Hongming Zhang, Haitao Mi, and Dong Yu. 2025b. Cognitive kernel-pro: A framework for deep research agents and agent foundation models training. Preprint, arXiv:2508.00414. — [arXiv:2508.00414](https://arxiv.org/abs/2508.00414) find
  5. Lang Feng, Zhenghai Xue, Tingcong Liu, and Bo An. 2025. Group-in-group policy optimization for llm agent training. Preprint, arXiv:2505.10978. find
  6. Huanang Gao, Jiayi Geng, Wenyue Hua, Mengkang Hu, Xinzhe Juan, Hongzhang Liu, Shilong Liu, Jiahao Qiu, Xuan Qi, Yiran Wu, Hongru Wang, Han Xiao, Yuhang Zhou, Shaokun Zhang, Jiayi Zhang, Jinyu Xiang, Yixiong Fang, Qiwen Zhao, Dongrui Liu, Qihan Ren, Cheng Qian, Zhenhailong Wang, Minda Hu, Huazheng Wa… find
  7. Miao Lu, Weiwei Sun, Weihua Du, Zhan Ling, Xuesong Yao, Kang Liu, and Jiecao Chen. 2025. Scaling llm multi-turn rl with end-to-end summarization-based context management. arXiv preprint arXiv:2510.06727. — [arXiv:2510.06727](https://arxiv.org/abs/2510.06727) find
  8. George A Miller. 1956. The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological review, 63(2):81. find
  9. Allen Newell, Herbert Alexander Simon, et al. 1972. Human problem solving, volume 104. Prentice-hall Englewood Cliffs, NJ. find
  10. Cheng Qian, Emre Can Acikgoz, Hongru Wang, Xiusi Chen, Avirup Sil, Dilek Hakkani-Tür, Gokhan Tur, and Heng Ji. 2025. SMART: Self-aware agent for tool overuse mitigation. In Findings of the Association for Computational Linguistics: ACL 2025, pages 4604–4621, Vienna, Austria. Association for Computat… find
  11. Jiahao Qiu, Xuan Qi, Tongcheng Zhang, Xinzhe Juan, Jiacheng Guo, Yifu Lu, Yimin Wang, Zixin Yao, Qihan Ren, Xun Jiang, Xing Zhou, Dongrui Liu, Ling Yang, Yue Wu, Kaixuan Huang, Shilong Liu, Hongru Wang, and Mengdi Wang. 2025. Alita: Generalist agent enabling scalable agentic reasoning with minimal p… — [arXiv:2505.20286](https://arxiv.org/abs/2505.20286) find
  12. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. Preprint, arXiv:1707.06347. find
  13. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  14. Junhong Shen, Hao Bai, Lunjun Zhang, Yifei Zhou, Amrith Setlur, Shengbang Tong, Diego Caples, Nan Jiang, Tong Zhang, Ameet Talwalkar, and Aviral Kumar. 2025. Thinking vs. doing: Agents that reason by scaling test-time interaction. Preprint, arXiv:2506.07976. — [arXiv:2506.07976](https://arxiv.org/abs/2506.07976) find
  15. Shijue Huang, Hongru WANG, Wanjun Zhong, Zhaochen Su, Jiazhan Feng, Bowen Cao, and Yi R. Fung. 2026. Adactrl: Towards adaptive and controllable reasoning via difficulty-aware budgeting. Transactions on Machine Learning Research. find
  16. Mohit Shridhar, Xingdi Yuan, Marc-Alexandre Côté, Yonatan Bisk, Adam Trischler, and Matthew Hausknecht. 2021. ALFWorld: Aligning Text and Embodied Environments for Interactive Learning. In Proceedings of the International Conference on Learning Representations (ICLR).
  17. Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan O Arik, Dong Wang, Hamed Zamani, and Jiawei Han. 2025. Search-r1: Training LLMs to reason and leverage search engines with reinforcement learning. In Second Conference on Language Modeling.
  18. Guangya Wan, Mingyang Ling, Xiaoqi Ren, Rujun Han, Sheng Li, and Zizhao Zhang. 2025. Compass: Enhancing agent long-horizon reasoning with evolving context. Preprint, arXiv:2510.08790.
  19. Mukai Li, Qingcheng Zeng, Tianqing Fang, Zhenwen Liang, Linfeng Song, Qi Liu, Haitao Mi, and Dong Yu. 2026. Verified critical step optimization for LLM agents. CoRR, abs/2602.03412. find
  20. Hongru Wang, Cheng Qian, Manling Li, Jiahao Qiu, Boyang Xue, Mengdi Wang, Heng Ji, Amos Storkey, and Kam-Fai Wong. 2026. Position: Agent should invoke external tools only when epistemically necessary. Preprint, arXiv:2506.00886. — [arXiv:2506.00886](https://arxiv.org/abs/2506.00886) find
  21. Hongru Wang, Cheng Qian, Wanjun Zhong, Xiusi Chen, Jiahao Qiu, Shijue Huang, Bowen Jin, Mengdi Wang, Kam-Fai Wong, and Heng Ji. 2025a. Acting less is reasoning more! teaching model to act efficiently. Preprint, arXiv:2504.14870. — [arXiv:2504.14870](https://arxiv.org/abs/2504.14870) find
  22. Aohan Zeng, Mingdao Liu, Rui Lu, Bowen Wang, Xiao Liu, Yuxiao Dong, and Jie Tang. 2024. Agenttuning: Enabling generalized agent abilities for llms. In Findings of the Association for Computational Linguistics: ACL 2024, pages 3053–3077.
  23. Hongru Wang, Rui Wang, Fei Mi, Yang Deng, Zezhong Wang, Bin Liang, Ruifeng Xu, and Kam-Fai Wong. 2023. Cue-CoT: Chain-of-thought prompting for responding to in-depth dialogue questions with LLMs. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 12047–12064, Singapore.… find
  24. Guibin Zhang, Hejia Geng, Xiaohang Yu, Zhenfei Yin, Zaibin Zhang, Zelin Tan, Heng Zhou, Zhongzhi Li, Xiangyuan Xue, Yijiang Li, Yifan Zhou, Yang Chen, Chen Zhang, Yutao Fan, Zihu Wang, Songtao Huang, Francisco Piedrahita-Velez, Yue Liao, Hongru Wang, Mengyue Yang, Heng Ji, Jun Wang, Shuicheng Yan, P… — [arXiv:2509.02547](https://arxiv.org/abs/2509.02547) find
  25. Rui Wang, Ce Zhang, Jun-Yu Ma, Jianshu Zhang, Hongru Wang, Yi Chen, Boyang Xue, Tianqing Fang, Zhisong Zhang, Hongming Zhang, Haitao Mi, Dong Yu, and Kam-Fai Wong. 2025b. Explore to evolve: Scaling evolved aggregation logic via proactive online exploration for deep research agents. Preprint, arXiv:2… — [arXiv:2510.14438](https://arxiv.org/abs/2510.14438) find
  26. Ruoyao Wang, Peter Jansen, Marc-Alexandre Côté, and Prithviraj Ammanabrolu. 2022. ScienceWorld: Is your agent smarter than a 5th grader? In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11279–11298, Abu Dhabi, United Arab Emirates. Association for Comp… find
  27. Xixi Wu, Kuan Li, Yida Zhao, Liwen Zhang, Litu Ou, Huifeng Yin, Zhongwang Zhang, Xinmiao Yu, Dingchu Zhang, Yong Jiang, et al. 2025. Resum: Unlocking long-horizon search intelligence via context summarization. arXiv preprint arXiv:2509.13313.
  28. Zijing Zhang, Ziyang Chen, Mingxiao Li, Zhaopeng Tu, and Xiaolong Li. 2025b. Rlvmr: Reinforcement learning with verifiable meta-reasoning rewards for robust long-horizon agents. Preprint, arXiv:2507.22844. — [arXiv:2507.22844](https://arxiv.org/abs/2507.22844) find
  29. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems, 36:46595–46623.
  30. Zijian Zhou, Ao Qu, Zhaoxuan Wu, Sunghwan Kim, Alok Prakash, Daniela Rus, Jinhua Zhao, Bryan Kian Hsiang Low, and Paul Pu Liang. 2025. Mem1: Learning to synergize memory and reasoning for efficient long-horizon agents. arXiv preprint arXiv:2506.15841. — [arXiv:2506.15841](https://arxiv.org/abs/2506.15841) find
  31. Zhiheng Xi, Yiwen Ding, Wenxiang Chen, Boyang Hong, Honglin Guo, Junzhe Wang, Dingwen Yang, Chenyang Liao, Xin Guo, Wei He, Songyang Gao, Lu Chen, Rui Zheng, Yicheng Zou, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing Huang, Zuxuan Wu, and Yu-Gang Jiang. 2024. Agentgym: Evolving large language model-based… find
  32. Shunyu Yao, Howard Chen, John Yang, and Karthik Narasimhan. 2022a. Webshop: Towards scalable real-world web interaction with grounded language agents. Advances in Neural Information Processing Systems, 35:20744–20757.
  33. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2022b. React: Synergizing reasoning and acting in language models. In The eleventh international conference on learning representations.
  34. Rui Ye, Zhongwang Zhang, Kuan Li, Huifeng Yin, Zhengwei Tao, Yida Zhao, Liangcai Su, Liwen Zhang, Zile Qiao, Xinyu Wang, Pengjun Xie, Fei Huang, Siheng Chen, Jingren Zhou, and Yong Jiang. 2025. Agentfold: Long-horizon web agents with proactive context management. Preprint, arXiv:2510.24699. — [arXiv:2510.24699](https://arxiv.org/abs/2510.24699) find
  35. Data \# trajs \# samples \# summaries \# assistant turns AlfWorld 300 978 678 4,051 SciWorld 300 1199 899 6,721 *Table 6: The data statistics analysis for both AlfWorld and SciWorld.* *Figure 6: Error analysis results.* find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-self-sum-teaching-agent-decide/