Elamur Research⌕
☰
home / papers / 2026 / COMPASS: Enhancing Agent Long-Horizon Reasoning with Evolving Context

COMPASS: Enhancing Agent Long-Horizon Reasoning with Evolving Context

Paper · 2026 · Guangya Wan, Mingyang Ling, Xiaoqi Ren et al.

Summary

LLM agents often fail on long-horizon tasks that need many chained tool calls. Small mistakes pile up, and as the history grows the agent misses key evidence, gets distracted, or stops too early instead of changing its plan. The authors argue that context management is the main obstacle. They propose COMPASS, a lightweight hierarchical framework with three roles:

  • a Main Agent that runs ReAct-style reasoning and tool use;
  • a Meta-Thinker that watches progress asynchronously and steps in with strategic decisions (persist, revise, stop);
  • a Context Manager that compresses the accumulated history into short structured briefs that refresh the Main Agent's context.

This keeps the role and context separation of multi-agent systems while staying as autonomous as a single agent, with no human-in-the-loop signals. On GAIA, BrowseComp and Humanity's Last Exam, COMPASS beats single- and multi-agent baselines by up to 20% relative accuracy. A test-time scaling variant reaches the level of established DeepResearch agents. A post-trained smaller model can take over context management to cut costs. The work offers practical design principles for scalable agents.

Key points

  • The paper separates tactical reasoning (carrying out the current plan step) from strategic reasoning (deciding when to correct, replan or finish). It defines long-horizon tasks as needing roughly more than 10 interdependent steps.
  • Evaluation covers GAIA, BrowseComp and Humanity's Last Exam, with gains of up to 20% relative to baselines. The experiments run Gemini 2.5 Pro and Flash, along with Gemma-3, all accessed through Google Cloud's Vertex AI platform.
  • COMPASS-TTS offers three parallel-sampling modes:
    • Full-PS gives the largest accuracy gains but the highest token cost.
    • MT-PS (meta-thinking only) is a middle option.
    • CM-PS (context manager only) is the most efficient.
  • On BrowseComp, accuracy plateaus near n = 4 samples, so n = 2–4 is the recommended range.
  • The released CONTEXT-12B model handles context management efficiently. Stated limitations: tests use only search, text browsing and code tools, and the study mostly relies on proprietary models.

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 VirginiaCharlottesville · United States36 works · 9,730 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.Reasoning · 1,426 works

Cite

Cite as Guangya Wan, Mingyang Ling, Xiaoqi Ren, Rujun Han, Sheng Li and Zizhao Zhang. 2026. COMPASS: Enhancing Agent Long-Horizon Reasoning with Evolving Context. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.152

@inproceedings{wan2026compass,
  title = {{COMPASS: Enhancing Agent Long-Horizon Reasoning with Evolving Context}},
  author = {Guangya Wan, Mingyang Ling, Xiaoqi Ren, Rujun Han, Sheng Li and Zizhao Zhang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.152},
  url = {https://doi.org/10.18653/v1/2026.acl-long.152}
}

References

53 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anthropic. 2024. Building effective agents. https://www.anthropic.com/engineering/ building-effective-agents. Accessed: 2025-09-24. ↗
  2. Anthropic. 2025. The “think” tool: Enabling claude to stop and think in complex tool use situations. https://www.anthropic.com/ engineering/claude-think-tool. Accessed: 2025-09-22. ↗
  3. Trishna Chakraborty, Udita Ghosh, Xiaopan Zhang, Fahim Faisal Niloy, Yue Dong, Jiachen Li, Amit K. Roy-Chowdhury, and Chengyu Song. 2025. Heal: An empirical study on hallucinations in embodied agents driven by large language models. Preprint, arXiv:2506.15065. find
  4. Qiguang Chen, Libo Qin, Jinhao Liu, Dengyun Peng, Jiannan Guan, Peng Wang, Mengkang Hu, Yuhang Zhou, Te Gao, and Wanxiang Che. 2025. Towards reasoning era: A survey of long chain-of-thought for reasoning large language models. Preprint, arXiv:2503.09567.
  5. Gheorghe et al. Comanici. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. Preprint, arXiv:2507.06261.
  6. Yiran Ding, Li Lyna Zhang, Chengruidong Zhang, Yuanyuan Xu, Ning Shang, Jiahang Xu, Fan Yang, and Mao Yang. 2024. Longrope: Extending llm context window beyond 2 million tokens. In Proceedings of the 41st International Conference on Machine Learning (ICML 2024).
  7. Lutfi Eren Erdogan, Hiroki Furuta, Sehoon Kim, Nicholas Lee, Suhong Moon, Gopala Anumanchipalli, Kurt Keutzer, and Amir Gholami. 2025. Plan-and-act: Improving planning of agents for long-horizon tasks. In Forty-second International Conference on Machine Learning. find
  8. DeepSeek-AI et al. 2025. DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning. Preprint, arXiv:2501.12948.
  9. Jiazhan Feng, Shijue Huang, Xingwei Qu, Ge Zhang, Yujia Qin, Baoquan Zhong, Chengquan Jiang, Jinxin Chi, and Wanjun Zhong. 2026. Retool: Reinforcement learning for strategic tool use in LLMs. In The Fourteenth International Conference on Learning Representations. find
  10. In Gim, Guojun Chen, Seung seob Lee, Nikhil Sarda, Anurag Khandelwal, and Lin Zhong. 2024. Prompt cache: Modular attention reuse for low-latency inference. Preprint, arXiv:2311.04934. find
  11. Gonzalo Gonzalez-Pumariega, Leong Su Yean, Neha Sunkara, and Sanjiban Choudhury. 2025. Robotouille: An asynchronous planning benchmark for llm agents. Preprint, arXiv:2502.05227. 3368 9 — [arXiv:2502.05227](https://arxiv.org/abs/2502.05227) find
  12. Rujun Han, Yanfei Chen, Zoey CuiZhu, Lesly Miculicich, Guan Sun, Yuanjun Bi, Weiming Wen, Hui Wan, Chunfeng Wen, Solène Maître, George Lee, Vishy Tirumalashetty, Emily Xue, Zizhao Zhang, Salem Haykal, Burak Gokturk, Tomas Pfister, and Chen-Yu Lee. 2025. Deep researcher with test-time diffusion. Prep… — [arXiv:2507.16075](https://arxiv.org/abs/2507.16075) find
  13. Kelly Hong, Anton Troynikov, and Jeff Huber. 2025. Context rot: How increasing input tokens impacts llm performance. Technical report, Chroma Research. Technical report. find
  14. Xinyi Hou, Yanjie Zhao, Shenao Wang, and Haoyu Wang. 2025. Model context protocol (mcp): Landscape, security threats, and future research directions. Preprint, arXiv:2503.23278. — [arXiv:2503.23278](https://arxiv.org/abs/2503.23278) find
  15. Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Xinying Song, and Denny Zhou. 2024a. Large language models cannot self-correct reasoning yet. In The Twelfth International Conference on Learning Representations.
  16. Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. 2024b. Understanding the planning of llm agents: A survey. Preprint, arXiv:2402.02716. find
  17. Yuxuan Huang, Yihang Chen, Haozheng Zhang, Kang Li, Huichi Zhou, Meng Fang, Linyi Yang, Xiaoguang Li, Lifeng Shang, Songcen Xu, Jianye Hao, Kun Shao, and Jun Wang. 2025. Deep research agents: A systematic examination and roadmap. Preprint, arXiv:2506.18096. — [arXiv:2506.18096](https://arxiv.org/abs/2506.18096) find
  18. Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan Arik, Dong Wang, Hamed Zamani, and Jiawei Han. 2025. Search-r1: Training llms to reason and leverage search engines with reinforcement learning. arXiv preprint arXiv:2503.09516.
  19. Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. 2023. CAMEL: Communicative agents for ”mind” exploration of large language model society. In Thirty-seventh Conference on Neural Information Processing Systems.
  20. Hang Li, Yucheng Chu, Kaiqi Yang, Yasemin Copur-Gencturk, and Jiliang Tang. 2025a. Llm-based automated grading with human-in-the-loop. Preprint, arXiv:2504.05239. — [arXiv:2504.05239](https://arxiv.org/abs/2504.05239) find
  21. Weizhen Li, Jianbo Lin, Zhuosong Jiang, Jingyi Cao, Xinpeng Liu, Jiayu Zhang, Zhenqiang Huang, Qianben Chen, Weichen Sun, Qiexiang Wang, Hongxuan Lu, Tianrui Qin, Chenghao Zhu, Yi Yao, Shuying Fan, Xiaowan Li, Tiannan Wang, Pai Liu, King Zhu, and 11 others. 2025b. Chain-of-agents: End-to-end agent f… — [arXiv:2508.13167](https://arxiv.org/abs/2508.13167) find
  22. Xiaomin Li, Zhou Yu, Zhiwei Zhang, Xupeng Chen, Ziji Zhang, Yingying Zhuang, Narayanan Sadagopan, and Anurag Beniwal. 2025c. When thinking fails: The pitfalls of reasoning for instruction-following in llms. Preprint, arXiv:2505.11423. — [arXiv:2505.11423](https://arxiv.org/abs/2505.11423) find
  23. Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Shuming Shi, and Zhaopeng Tu. 2024. Encouraging divergent thinking in large language models through multi-agent debate. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 17…
  24. Jack et al. Lindsey. 2025. On the biology of a large language model. https://transformer-circuits.pub/2025/attribution-graphs/biology.html. Transformer Circuits; published March 27, 2025; accessed 2025-09-10. ↗
  25. Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024. Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics, 12:157–173.
  26. Zichen et al. Liu. 2025. There may not be aha moment in r1-zero-like training — a pilot study. https://oatllm.notion.site/oat-zero. Notion blog; accessed 2025-09-10. ↗
  27. 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
  28. Lingrui Mei, Jiayu Yao, Yuyao Ge, Yiwei Wang, Baolong Bi, Yujun Cai, Jiazhi Liu, Mingyu Li, Zhong-Zhi Li, Duzhen Zhang, Chenlin Zhou, Jiayi Mao, Tianze Xia, Jiafeng Guo, and Shenghua Liu. 2025. A survey of context engineering for large language models. Preprint, arXiv:2507.13334. — [arXiv:2507.13334](https://arxiv.org/abs/2507.13334) find
  29. Grégoire Mialon, Clémentine Fourrier, Thomas Wolf, Yann LeCun, and Thomas Scialom. 2024. GAIA: a benchmark for general AI assistants. In The Twelfth International Conference on Learning Representations.
  30. Shervin Minaee, Tomas Mikolov, Narjes Nikzad, Meysam Chenaghlu, Richard Socher, Xavier Amatriain, and Jianfeng Gao. 2025. Large language models: A survey. Preprint, arXiv:2402.06196.
  31. Melissa Z Pan, Mert Cemri, Lakshya A Agrawal, Shuyi Yang, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Kannan Ramchandran, Dan Klein, Joseph E. Gonzalez, Matei Zaharia, and Ion Stoica. 2025. Why do multiagent systems fail? In ICLR 2025 Workshop on Building Trust in Language Mode… find
  32. Mihir Parmar, Palash Goyal, Xin Liu, Yiwen Song, Mingyang Ling, Chitta Baral, Hamid Palangi, and Tomas Pfister. 2025. Plan-tuning: Post-training language models to learn step-by-step planning for complex problem solving. Preprint, arXiv:2507.07495. find
  33. Long et al. Phan. 2025. Humanity’s Last Exam. Preprint, arXiv:2501.14249.
  34. Aske et al. Plaat. 2025. Survey on agentic large language models: reasoning, acting, and interacting. ACM Computing Surveys. find
  35. Rafael et al. Rafailov. 2024. Direct preference optimization: Your language model is secretly a reward model. arXiv preprint arXiv:2305.18290.
  36. Ranjan Sapkota, Konstantinos I. Roumeliotis, and Manoj Karkee. 2025. Ai agents vs. agentic ai: A conceptual taxonomy, applications and challenges. Information Fusion, 126:103599. find
  37. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik R Narasimhan, and Shunyu Yao. 2023. Reflexion: language agents with verbal reinforcement learning. In Thirty-seventh Conference on Neural Information Processing Systems.
  38. Charlie Victor Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2025. Scaling LLM test-time compute optimally can be more effective than scaling parameters for reasoning. In The Thirteenth International Conference on Learning Representations. find
  39. Haotian Sun, Yuchen Zhuang, Lingkai Kong, Bo Dai, and Chao Zhang. 2023. Adaplanner: Adaptive planning from feedback with language models. In Thirty-seventh Conference on Neural Information Processing Systems. find
  40. Wannita Takerngsaksiri, Jirat Pasuksmit, Patanamon Thongtanunam, Chakkrit Tantithamthavorn, Ruix-iong Zhang, Fan Jiang, Jing Li, Evan Cook, Kun Chen, and Ming Wu. 2025. Human-in-the-loop software development agents. In 2025 IEEE/ACM 47th International Conference on Software Engineering: Software Eng… find
  41. Khanh-Tung Tran, Dung Dao, Minh-Duong Nguyen, Quoc-Viet Pham, Barry O’Sullivan, and Hoang D. Nguyen. 2025. Multi-agent collaboration mechanisms: A survey of llms. Preprint, arXiv:2501.06322. find
  42. Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. 2024. Voyager: An open-ended embodied agent with large language models. Transactions on Machine Learning Research. find
  43. Haozhe Wang, Qixin Xu, Che Liu, Junhong Wu, Fangzhen Lin, and Wenhu Chen. 2025. Emergent hierarchical reasoning in llms through reinforcement learning. Preprint, arXiv:2509.03646. — [arXiv:2509.03646](https://arxiv.org/abs/2509.03646) find
  44. Lei Wang, Wanyu Xu, Yihuai Lan, Zhiqiang Hu, Yunshi Lan, Roy Ka-Wei Lee, and Ee-Peng Lim. 2023a. Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Lo… find
  45. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023b. Self-consistency improves chain of thought reasoning in language models. In The Eleventh International Conference on Learning Representations.
  46. Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Isa Fulford, Hyung Won Chung, Alex Tachard Passos, William Fedus, and Amelia Glaese. 2025. Browsecomp: A simple yet challenging benchmark for browsing agents. Preprint, arXiv:2504.12516. find
  47. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed H. Chi, Quoc V Le, and Denny Zhou. 2022. Chain of thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems.
  48. Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan Awadallah, Ryen W White, Doug Burger, and Chi Wang. 2024. Autogen: Enabling next-gen LLM applications via multi-agent conversations. In First Conference on Language… find
  49. Zhiheng Xi, Jixuan Huang, Chenyang Liao, Baodai Huang, Honglin Guo, Jiaqi Liu, Rui Zheng, Junjie Ye, Jiazheng Zhang, Wenxiang Chen, Wei He, Yiwen Ding, Guanyu Li, Zehui Chen, Zhengyin Du, Xuesong Yao, Yufei Xu, Jiecao Chen, Tao Gui, and 4 others. 2025. Agentgym-rl: Training llm agents for long-horiz… — [arXiv:2509.08755](https://arxiv.org/abs/2509.08755) find
  50. Ziwei Xu, Sanjay Jain, and Mohan Kankanhalli. 2025. Hallucination is inevitable: An innate limitation of large language models. Preprint, arXiv:2401.11817.
  51. 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).
  52. Guibin Zhang, Hejia Geng, Xiaohang Yu, Zhenfei Yin, Zelin Tan, Heng Zhou, Zhongzhi Li, Xiangyuan Xue, Yijiang Li, Yifan Zhou, Yang Chen, Chen Zhang, Yutao Fan, Zihu Wang, Songtao Huang, Yue Liao, Hongru Wang, Mengyue Yang, Heng Ji, and 5 others. 2025. The landscape of agentic reinforcement learning… — [arXiv:2509.02547](https://arxiv.org/abs/2509.02547) find
  53. Muru Zhang, Ofir Press, William Merrill, Alisa Liu, and Noah A. Smith. 2023. How language model hallucinations can snowball. Preprint, arXiv:2305.13534.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wan-2026-compass-enhancing-agent-long-horizon/