Elamur Research⌕
☰
home / papers / 2026 / ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents

ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents

Paper · 2026 · Yilun Yao, Shan Huang, Elsie Dai et al.

Summary

LLM-based research agents that run long search sessions tend to get worse as their interaction history grows, a problem called context rot. Common remedies either keep appending the full history or periodically compress it. Both treat context as a fixed record, so early mistakes or wrong priorities stay in place. ARC instead treats the agent's working context as an internal state that is maintained and corrected while the task runs. It separates two roles. An actor produces reasoning and actions. A dedicated, trainable Context Manager updates a compact memory after every turn and, only when it detects drift or errors, runs a reflection step that reorganizes the stored state. The Context Manager is trained with supervised fine-tuning on curated trajectories. On five long-horizon information-seeking benchmarks, ARC beats both raw-history and summarization baselines, with gains of up to 11 absolute points on BrowseComp-ZH. The results suggest that how an agent maintains its context can matter as much as the underlying model's raw reasoning ability.

Key points

  • The agent's state has three parts: a mutable Checklist of goals and progress, an Interaction Memory (IM) with condensed evidence, and the full Previous Interaction (reasoning, action, observation).

  • The Context Manager summarizes incrementally at every turn. Reflection is triggered only selectively and can rewrite both the checklist and the memory.

  • The largest reported gain is about +11% absolute on BrowseComp-ZH with Qwen2.5-32B-Instruct. Once trained, the Context Manager can beat untrained models that are considerably bigger.

  • Runtime was measured on 100 BrowseComp instances, using a Qwen3-32B actor and a 14B Context Manager:

    SetupTime per stepRelative to ReAct
    ReAct9.457 s1.00×
    ARC, sequential20.476 s2.17×
    ARC, overlapped (Context Manager runs in parallel with the next actor step)10.971 s1.16×

    Overlapping cuts ARC's step time by roughly 46%.

  • Code is published at github.com/Qihoo360/ARC-Context. Stated limitations are extra system overhead, evaluation limited to search tasks, and dependence on costly curated training trajectories.

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

🎓Peking UniversityBeijing · China315 works · 43,268 citations🎓Beihang UniversityBeijing · China86 works · 9,097 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

Cite

Cite as Yilun Yao, Shan Huang, Elsie Dai, Zhewen Tan, Zhenyu Duan, Shousheng Jia, Yanbing Jiang and Tong Yang. 2026. ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents. 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.930

@inproceedings{yao2026arc,
  title = {{ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents}},
  author = {Yilun Yao, Shan Huang, Elsie Dai, Zhewen Tan, Zhenyu Duan, Shousheng Jia, Yanbing Jiang and Tong Yang},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.930},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.930}
}

References

37 references, 18 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Danqi Chen, Adam Fisch, Jason Weston, and Antoine Bordes. 2017. Reading Wikipedia to answer open-domain questions. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1870–1879, Vancouver, Canada. Association for Computational Lin…
  2. Kaiyuan Chen, Yixin Ren, Yang Liu, Xiaobo Hu, Haotong Tian, Tianbao Xie, Fangfu Liu, Haoye Zhang, Hongzhang Liu, Yuan Gong, Chen Sun, Han Hou, Hui Yang, James Pan, Jianan Lou, Jiayi Mao, Jizheng Liu, Jinpeng Li, Kangyi Liu, and 14 others. 2025. xbench: Tracking agents productivity scaling with profe… — [arXiv:2506.13651](https://arxiv.org/abs/2506.13651) find
  3. Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. 2025. Mem0: Building production-ready ai agents with scalable long-term memory. Preprint, arXiv:2504.19413.
  4. Rajarshi Das, Shehzaad Dhuliawala, Manzil Zaheer, and Andrew McCallum. 2019. Multi-step retriever-reader interaction for scalable open-domain question answering. In International Conference on Learning Representations.
  5. DeepSeek-AI. 2025. Deepseek-v3.2: Pushing the frontier of open large language models.
  6. Johan Ferret, Raphaël Marinier, Matthieu Geist, and Olivier Pietquin. 2019. Self-attentional credit assignment for transfer in reinforcement learning. arXiv preprint arXiv:1907.08027. — [arXiv:1907.08027](https://arxiv.org/abs/1907.08027) find
  7. Zhibin Gou, Zhihong Shao, Yeyun Gong, Yelong Shen, Yujiu Yang, Nan Duan, and Weizhu Chen. 2024. Critic: Large language models can self-correct with tool-interactive critiquing. Preprint, arXiv:2305.11738.
  8. Xanh Ho, Anh-Khoa Duong Nguyen, Saku Sugawara, and Akiko Aizawa. 2020. Constructing a multi-hop QA dataset for comprehensive evaluation of reasoning steps. In Proceedings of the 28th International Conference on Computational Linguistics, pages 6609–6625, Barcelona, Spain (Online). International Comm…
  9. Kelly Hong, Anton Troynikov, and Jeff Huber. 2025. Context rot: How increasing input tokens impacts llm performance. Technical report, Chroma. find
  10. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-refine: Iterative refinement with… find
  11. Meta. 2025. The llama 4 herd: Llama 4 scout and maverick with unprecedented context length. https://ai.meta.com/blog/ llama-4-multimodal-intelligence/. ↗
  12. Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. 2023. Gaia: a benchmark for general ai assistants. Preprint, arXiv:2311.12983.
  13. OpenAI. 2025. Deep research: System card. https://cdn.openai.com/deep-research-system-card. pdf. System card. ↗
  14. OpenAI. 2025a. gpt-oss-120b and gpt-oss-20b model card. Preprint, arXiv:2508.10925.
  15. OpenAI. 2025b. Introducing gpt-4.1 in the api. https://openai.com/index/gpt-4-1/. ↗
  16. Siru Ouyang, Jun Yan, I-Hung Hsu, Yanfei Chen, Ke Jiang, Zifeng Wang, Rujun Han, Long T. Le, Samira Daruki, Xiangru Tang, Vishy Tirumalashetty, George Lee, Mahsan Rofouei, Hangfei Lin, Jiawei Han, Chen-Yu Lee, and Tomas Pfister. 2025. Reasoningbank: Scaling agent self-evolving with reasoning memory.… — [arXiv:2509.25140](https://arxiv.org/abs/2509.25140) find
  17. Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, and Joseph E. Gonzalez. 2024. Memgpt: Towards llms as operating systems. Preprint, arXiv:2310.08560. find
  18. Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah A. Smith, and Mike Lewis. 2023. Measuring and narrowing the compositionality gap in language models. Preprint, arXiv:2210.03350. find
  19. Qwen Team. 2024. Qwen2.5: A party of foundation models. find
  20. Qwen Team. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388.
  21. Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. Preprint, arXiv:2303.11366.
  22. Naftali Tishby, Fernando C. Pereira, and William Bialek. 2000. The information bottleneck method. Preprint, arXiv:physics/0004057. — [arXiv:physics/0004057](https://arxiv.org/abs/physics/0004057) find
  23. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2022. MuSiQue: Multihop questions via single-hop question composition. Transactions of the Association for Computational Linguistics.
  24. Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Isa Fulford, Hyung Won Chung, Alex Tachard Passos, William Fedus, and Amelia Glaese. 2025a. Browsecomp: A simple yet challenging benchmark for browsing agents. Preprint, arXiv:2504.12516.
  25. Tianxin Wei, Noveen Sachdeva, Benjamin Coleman, Zhankui He, Yuanchen Bei, Xuying Ning, Mengting Ai, Yunzhe Li, Jingrui He, Ed H. Chi, Chi Wang, Shuo Chen, Fernando Pereira, Wang-Cheng Kang, and Derek Zhiyuan Cheng. 2025b. Evo-memory: Benchmarking llm agent test-time learning with self-evolving memor… — [arXiv:2511.20857](https://arxiv.org/abs/2511.20857) find
  26. Jialong Wu, Wenbiao Yin, Yong Jiang, Zhenglin Wang, Zekun Xi, Runnan Fang, Linhai Zhang, Yulan He, Deyu Zhou, Pengjun Xie, and Fei Huang. 2025a. WebWalker: Benchmarking LLMs in web traversal. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long P…
  27. Xixi Wu, Kuan Li, Yida Zhao, Liwen Zhang, Litu Ou, Huifeng Yin, Zhongwang Zhang, Xinmiao Yu, Dingchu Zhang, Yong Jiang, Pengjun Xie, Fei Huang, Minhao Cheng, Shuai Wang, Hong Cheng, and Jingren Zhou. 2025b. Resum: Unlocking long-horizon search intelligence via context summarization. Preprint, arXiv:…
  28. Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. 2025. A-mem: Agentic memory for llm agents. Preprint, arXiv:2502.12110.
  29. Sikuan Yan, Xiufeng Yang, Zuchao Huang, Ercong Nie, Zifeng Ding, Zonggen Li, Xiaowen Ma, Kristian Kersting, Jeff Z. Pan, Hinrich Schütze, Volker Tresp, and Yunpu Ma. 2025. Memory-r1: Enhancing large language model agents to manage and utilize memories via reinforcement learning. Preprint, arXiv:2508…
  30. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W. Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Conference on Empirical Methods in Natural Language Processing (EMNLP).
  31. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR).
  32. 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
  33. Guibin Zhang, Haotian Ren, Chong Zhan, Zhenhong Zhou, Junhao Wang, He Zhu, Wangchunshu Zhou, and Shuicheng Yan. 2025a. Memevolve: Meta-evolution of agent memory systems. Preprint, arXiv:2512.18746. — [arXiv:2512.18746](https://arxiv.org/abs/2512.18746) find
  34. Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, Fenglu Hong, Vamsidhar Kamanuru, Jay Rainton, Chen Wu, Mengmeng Ji, Hanchen Li, Urmish Thakker, James Zou, and Kunle Olukotun. 2025b. Agentic context engineering: Evolving contexts for self-improving language models. Preprint, arXiv:2510.0461… find
  35. Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2023. Memorybank: Enhancing large language models with long-term memory. Preprint, arXiv:2305.10250. find
  36. Peilin Zhou, Bruce Leon, Xiang Ying, Can Zhang, Yifan Shao, Qichen Ye, Dading Chong, Zhiling Jin, Chenxuan Xie, Meng Cao, Yuxin Gu, Sixin Hong, Jing Ren, Jian Chen, Chao Liu, and Yining Hua. 2025a. Browsecomp-zh: Benchmarking web browsing ability of large language models in chinese. Preprint, arXiv:… — [arXiv:2504.19314](https://arxiv.org/abs/2504.19314) find
  37. Yuqi Zhou, Sunhao Dai, Zhanshuo Cao, Xiao Zhang, and Jun Xu. 2025b. Length-induced embedding collapse in plm-based models. Preprint, arXiv:2410.24200. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yao-2026-arc-active-reflection-driven-context/