Elamur Research⌕
☰
home / papers / 2025 / ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization

ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization

Paper · 2025 · Xixi Wu, Kuan Li, Yida Zhao et al.

Summary

Web agents built on LLMs often fail on hard multi-hop questions. The ReAct loop keeps adding every search result to the prompt, so the context window fills up before the agent finds the answer. Earlier fixes such as MEM1 or MemAgent change the model's architecture and require full retraining. ReSum avoids this with a workflow-level change. When the context grows too large, an external summarizer turns the accumulated history into a short, goal-focused digest. The agent then continues from the original question plus that digest, so exploration is no longer capped by window size and existing agents stay compatible.

The authors make two further contributions:

  • ReSumTool-30B, a dedicated summarizer distilled from a stronger teacher model.
  • ReSum-GRPO, an RL variant of GRPO. It splits each trajectory at its summary points and assigns the same final-outcome advantage to every piece, which addresses long-horizon credit assignment.

Without any training, ReSum beats ReAct by 4.5% on average across three benchmarks. RL adaptation adds another 8.2%. This shows that cheap context compression can substantially extend what search agents can solve.

Key points

  • ReSumTool-30B: Qwen3-30B-A3B-Thinking fine-tuned on ⟨Conversation, Summary⟩ pairs, with GPT-OSS-120B as the teacher. It is reported to produce better summaries than DeepSeek-R1-671B.
  • Data efficiency: ReSum-GRPO uses only 1K training samples. With it, WebSailor-30B reaches 33.3% Pass@1 on BrowseComp-zh. This beats ASearcher-32B (15.6%), MiroThinker-32B (17.0%) and WebExplorer-8B (32.0%), all trained on 10K+ samples. WebSailor-3B rises from 8.2% to 20.5%.
  • Extensive-context agents: on Tongyi-DeepResearch-30B-A3B, ReSum outperforms ReAct at every context limit from 32k to 128k. The biggest gains come under tighter budgets.
  • Costs: RL training takes roughly 1.5× longer. MEM1-GRPO edges ahead on BrowseComp by 1.2%, but it uses about 3× more tokens than ReSum.
  • Tools: the agent uses Google Search (top-10 results per query) and page visits via Jina. Evidence is extracted from pages by Qwen2.5-72B-Instruct. Summarization is triggered by rules, such as hitting a token limit, rather than by the agent itself.

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

 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.Summarization · 232 works

Cite

Cite as 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. 2025. ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization. arXiv:2509.13313.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2509.13313 · arXiv 2509.13313

@misc{wu2025resum,
  title = {{ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization}},
  author = {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},
  year = {2025},
  doi = {10.48550/arxiv.2509.13313},
  eprint = {2509.13313},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2509.13313}
}

References

38 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Chae, H., Kim, N., iunn Ong, K. T., Gwak, M., Song, G., Kim, J., Kim, S., Lee, D., and Yeo, J. Web agents with world models: Learning and leveraging environment dynamics in web navigation. In The Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum?… ↗
  2. Li, K., Zhang, Z., Yin, H., Zhang, L., Ou, L., Wu, J., Yin, W., Li, B., Tao, Z., Wang, X., Shen, W., Zhang, J., Zhang, D., Wu, X., Jiang, Y., Yan, M., Xie, P., Huang, F., and Zhou, J. Websailor: Navigating super-human reasoning for web agent. arXiv preprint arXiv:2507.02592, 2025a. find
  3. Li, W., Lin, J., Jiang, Z., Cao, J., Liu, X., Zhang, J., Huang, Christiano, P. F., Leike, J., Brown, T. B., Martic, M., Legg, Z., Chen, Q., Sun, W., Wang, Q., Lu, H., Qin, T., Zhu, S., and Amodei, D. Deep reinforcement learning from C., Yao, Y., Fan, S., Li, X., Wang, T., Liu, P., Zhu, K., human pre… — [arXiv:2508.13167](https://arxiv.org/abs/2508.13167) find
  4. Li, Z., Song, S., Xi, C., Wang, H., Tang, C., Niu, S., Chen, D., Yang, J., Li, C., Yu, Q., Zhao, J., Wang, Y., Liu, P., Lin, Z., Wang, P., Huo, J., Chen, T., Chen, K., Li, K., Tao, Z., Ren, J., Lai, H., Wu, H., Tang, B., Wang, Z., Fan, Z., Zhang, N., Zhang, L., Yan, J., Yang, M., Xu, T., Xu, W., Che… — [arXiv:2507.03724](https://arxiv.org/abs/2507.03724) find
  5. Dong, G., Mao, H., Ma, K., Bao, L., Chen, Y., Wang, Z., Chen, Z., Du, J., Wang, H., Zhang, F., et al. Agentic reinforced policy optimization. arXiv preprint arXiv:2507.19849, 2025. — [arXiv:2507.19849](https://arxiv.org/abs/2507.19849) find
  6. Gao, J., Fu, W., Xie, M., Xu, S., He, C., Mei, Z., Zhu, B., and Wu, Y. Beyond ten turns: Unlocking long-horizon agentic search with large-scale asynchronous rl, 2025. URL https://arxiv.org/abs/2508.07976. — [arXiv:2508.07976](https://arxiv.org/abs/2508.07976) ↗
  7. Liu, J., Li, Y., Zhang, C., Li, J., Chen, A., Ji, K., Cheng, W., Wu, Z., Du, C., Xu, Q., Song, J., Zhu, Z., Chen, W., Zhao, P., and He, J. Webexplorer: Explore and evolve for training long-horizon web agents, 2025. URL https://arxiv.org/abs/2509.06501. — [arXiv:2509.06501](https://arxiv.org/abs/2509.06501) ↗
  8. Gu, J., Jiang, X., Shi, Z., Tan, H., Zhai, X., Xu, C., Li, W., Shen, Y., Ma, S., Liu, H., et al. A survey on llm-as-a-judge. arXiv preprint arXiv:2411.15594, 2024. find
  9. Guo, D., Yang, D., Zhang, H., Song, J., Zhang, R., Xu, R., Zhu, Q., Ma, S., Wang, P., Bi, X., et al. DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025.
  10. Lu, R., Hou, Z., Wang, Z., Zhang, H., Liu, X., Li, Y., Feng, S., Tang, J., and Dong, Y. Deepdive: Advancing deep search agents with knowledge graphs and multi-turn rl, 2025. URL https://arxiv.org/abs/2509. 10446. ↗
  11. Jiang, A. Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D. S., de Las Casas, D., Bressand, F., Lengyel, Mialon, G., Fourrier, C., Wolf, T., LeCun, Y., and Scialom, T. Gaia: a benchmark for general ai assistants. In The Twelfth International Conference on Learning Representations, 2023.
  12. Team, Q. Qwen3 technical report, 2025. URL https://arxiv.org/abs/2505.09388. ↗
  13. Team, T. D., Li, B., Zhang, B., Zhang, D., Huang, F., Li, G., Chen, G., Yin, H., Wu, J., Zhou, J., et al. Tongyi deepresearch technical report. arXiv preprint arXiv:2510.24701, 2025c. — [arXiv:2510.24701](https://arxiv.org/abs/2510.24701) find
  14. OpenAI. gpt-oss-120b & gpt-oss-20b model card, 2025a. URL https://cdn.openai.com/pdf/ 419b6906-9da6-406c-a19d-1bb078ac7637/ oai_gpt-oss_model_card.pdf. ↗
  15. Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., Chen, Z., Tang, J., Chen, X., Lin, Y., et al. A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6):186345, 2024.
  16. OpenAI. Introducing openai o3 and o4-mini, 2025b. URL https://openai.com/index/ introducing-o3-and-o4-mini/. ↗
  17. OpenAI. Openai deep research. https://openai.com/index/introducing-deep-research/, 2025. ↗
  18. Wei, J., Karina, N., Chung, H. W., Jiao, Y. J., Papay, S., Glaese, A., Schulman, J., and Fedus, W. Measuring short-form factuality in large language models, 2024. URL https://arxiv.org/abs/2411.04368. ↗
  19. Wei, J., Sun, Z., Papay, S., McKinney, S., Han, J., Fulford, I., Chung, H. W., Passos, A. T., Fedus, W., and Glaese, A. Browsecomp: A simple yet challenging benchmark for browsing agents. arXiv preprint arXiv:2504.12516, 2025. find
  20. Ouyang, S., Yan, J., Hsu, I.-H., Chen, Y., Jiang, K., Wang, Z., Han, R., Le, L. T., Daruki, S., Tang, X., Tirumalashetty, V., Lee, G., Rofouei, M., Lin, H., Han, J., Lee, C.-Y., and Pfister, T. Reasoningbank: Scaling agent self-evolving with reasoning memory, 2025. URL https://arxiv.org/abs/2509.251… — [arXiv:2509.25140](https://arxiv.org/abs/2509.25140) ↗
  21. Wu, J., Li, B., Fang, R., Yin, W., Zhang, L., Tao, Z., Zhang, D., Xi, Z., Jiang, Y., Xie, P., Huang, F., and Zhou, J. Webdancer: Towards autonomous information seeking agency. arXiv preprint arXiv:2505.22648, 2025a.
  22. Schulman, J., Wolski, F., Dhariwal, P., Radford, A., and Klimov, O. Proximal policy optimization algorithms. CoRR, abs/1707.06347, 2017. find
  23. Wu, J., Yin, W., Jiang, Y., Wang, Z., Xi, Z., Fang, R., Zhou, D., Xie, P., and Huang, F. Webwalker: Benchmarking llms in web traversal. arXiv preprint arXiv:2501.07572, 2025b.
  24. Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M., Li, Y., Wu, Y., et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024.
  25. Xbench-Team. Xbench-deepsearch, 2025. URL https://xbench.org/agi/aisearch. ↗
  26. Tan, S., Luo, M., Cai, C., Venkat, T., Montgomery, K., Hao, A., Wu, T., Balyan, A., Roongta, M., Wang, C., Li, L. E., Popa, R. A., and Stoica, I. rllm: A framework for post-training language agents, 2025. Notion Blog. find
  27. Xi, Z., Chen, W., Guo, X., He, W., Ding, Y., Hong, B., Zhang, M., Wang, J., Jin, S., Zhou, E., et al. The rise and potential of large language model based agents: A survey. Science China Information Sciences, 68(2):121101, 2025.
  28. Tao, Z., Wu, J., Yin, W., Zhang, J., Li, B., Shen, H., Li, K., Zhang, L., Wang, X., Jiang, Y., Xie, P., Huang, F., and Zhou, J. Webshaper: Agentically data synthesizing via information-seeking formalization. arXiv preprint arXiv:2507.15061, 2025. — [arXiv:2507.15061](https://arxiv.org/abs/2507.15061) find
  29. Xu, W., Liang, Z., Mei, K., Gao, H., Tan, J., and Zhang, Y. A-mem: Agentic memory for llm agents. arXiv preprint arXiv:2502.12110, 2025.
  30. Yang, A., Yang, B., Zhang, B., Hui, B., Zheng, B., Yu, B., Li, C., Liu, D., Huang, F., Wei, H., et al. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115, 2024. find
  31. Team, K., Bai, Y., Bao, Y., Chen, G., Chen, J., Chen, N., Chen, R., Chen, Y., Chen, Y., Chen, Y., Chen, Z., Cui, J., Ding, H., Dong, M., Du, A., Du, C., Du, D., Du, Y., Fan, Y., Feng, Y., Fu, K., Gao, B., Gao, H., Gao, P., and et al. Kimi k2: Open agentic intelligence, 2025a. URL https://arxiv.org/a… ↗
  32. Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., and Cao, Y. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR), 2023.
  33. Team, M., Bai, S., Bing, L., Chen, C., Chen, G., Chen, Y., Chen, Z., Chen, Z., Dong, X., et al. Mirothinker: Pushing the performance boundaries of open-source research agents via model, context, and interactive scaling. arXiv preprint arXiv:2511.11793, 2025b. — [arXiv:2511.11793](https://arxiv.org/abs/2511.11793) find
  34. Yu, H., Chen, T., Feng, J., Chen, J., Dai, W., Yu, Q., Zhang, Y.-Q., Ma, W.-Y., Liu, J., Wang, M., et al. Memagent: Reshaping long-context llm with multi-conv rl-based memory agent. arXiv preprint arXiv:2507.02259, 2025a. — [arXiv:2507.02259](https://arxiv.org/abs/2507.02259) find
  35. Yu, Q., Zhang, Z., Zhu, R., Yuan, Y., Zuo, X., Yue, Y., Fan, T., Liu, G., Liu, L., Liu, X., et al. Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476, 2025b.
  36. Yuan, Q., Lou, J., Li, Z., Chen, J., Lu, Y., Lin, H., Sun, L., Zhang, D., and Han, X. Memsearcher: Training llms to reason, search and manage memory via end-to-end reinforcement learning, 2025. URL https://arxiv.org/abs/2511.02805. ↗
  37. Zhou, P., Leon, B., Ying, X., Zhang, C., Shao, Y., Ye, Q., Chong, D., Jin, Z., Xie, C., Cao, M., et al. Browsecomp-zh: Benchmarking web browsing ability of large language models in chinese. arXiv preprint arXiv:2504.19314, 2025a. — [arXiv:2504.19314](https://arxiv.org/abs/2504.19314) find
  38. Zhou, Z., Qu, A., Wu, Z., Kim, S., Prakash, A., Rus, D., Zhao, J., Low, B. K. H., and Liang, P. P. Mem1: Learning to synergize memory and reasoning for efficient long-horizon agents, 2025b. URL https://arxiv.org/ abs/2506.15841. — [arXiv:2506.15841](https://arxiv.org/abs/2506.15841) ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wu-2025-resum-unlocking-long-horizon-search/