![[2602.14234] REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14234] REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents
Summary
The paper presents REDSearcher, a novel framework designed to optimize long-horizon search agents by addressing the challenges of task synthesis and interaction costs, achieving state-of-the-art performance in both text-only and multimodal benchmarks.
Why It Matters
As large language models evolve into practical problem solvers, optimizing them for complex search tasks is crucial. REDSearcher offers a scalable and efficient approach to enhance the performance of search agents, making it relevant for researchers and developers in AI and machine learning.
Key Takeaways
- REDSearcher improves task synthesis through dual-constrained optimization.
- The framework encourages proactive tool use with tool-augmented queries.
- Core capabilities are strengthened during midtraining, reducing trajectory collection costs.
- A local simulated environment allows for rapid algorithmic iteration.
- The authors will release extensive datasets and code to support future research.
Computer Science > Artificial Intelligence arXiv:2602.14234 (cs) [Submitted on 15 Feb 2026] Title:REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents Authors:Zheng Chu, Xiao Wang, Jack Hong, Huiming Fan, Yuqi Huang, Yue Yang, Guohai Xu, Chenxiao Zhao, Cheng Xiang, Shengchao Hu, Dongdong Kuang, Ming Liu, Bing Qin, Xing Yu View a PDF of the paper titled REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents, by Zheng Chu and 13 other authors View PDF HTML (experimental) Abstract:Large language models are transitioning from generalpurpose knowledge engines to realworld problem solvers, yet optimizing them for deep search tasks remains challenging. The central bottleneck lies in the extreme sparsity of highquality search trajectories and reward signals, arising from the difficulty of scalable longhorizon task construction and the high cost of interactionheavy rollouts involving external tool calls. To address these challenges, we propose REDSearcher, a unified framework that codesigns complex task synthesis, midtraining, and posttraining for scalable searchagent optimization. Specifically, REDSearcher introduces the following improvements: (1) We frame task synthesis as a dualconstrained optimization, where task difficulty is precisely governed by graph topology and evidence dispersion, allowing scalable generation of complex, highquality tasks. (2) We introduce toolaugmented queries to encourage proactive tool use r...
