Search More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization
Qianben Chen, Tianrui Qin, King Zhu, Qiexiang Wang, Chengjun Yu, Shu Xu, Jiaqi Wu, Jiayu Zhang, Xinpeng Liu, Xin Gui, Jingyi Cao, Piaohong Wang, Dingfeng Shi, He Zhu, Tiannan Wang, Yuqing Wang, Maojia Song, Tianyu Zheng, Ge Zhang, Jian Yang, Jiaheng Liu, Minghao Liu
2026-02-27
Summary
This paper introduces a new approach to building AI agents that perform complex research tasks, aiming for both speed and the ability to handle different kinds of research questions.
What's the problem?
Current AI agents designed for research often try to solve problems by thinking through them step-by-step, which takes a lot of computing power and time, especially when they need to search for information. Also, these agents struggle to adapt to different research scenarios – what works well for answering simple questions might not work for more open-ended investigations.
What's the solution?
The researchers developed a framework called Search More, Think Less (SMTL). Instead of lengthy reasoning chains, SMTL focuses on quickly gathering relevant evidence in parallel. It also includes a system for creating diverse research tasks to train the agent, covering both straightforward questions and more complex, exploratory research. The agent is trained using a combination of showing it examples and letting it learn through trial and error.
Why it matters?
This work is important because it makes AI research agents more efficient and versatile. By reducing the amount of 'thinking' needed and improving their ability to generalize, these agents can tackle a wider range of research problems faster and more effectively, achieving impressive results on several standard benchmarks.
Abstract
Recent deep research agents primarily improve performance by scaling reasoning depth, but this leads to high inference cost and latency in search-intensive scenarios. Moreover, generalization across heterogeneous research settings remains challenging. In this work, we propose Search More, Think Less (SMTL), a framework for long-horizon agentic search that targets both efficiency and generalization. SMTL replaces sequential reasoning with parallel evidence acquisition, enabling efficient context management under constrained context budgets. To support generalization across task types, we further introduce a unified data synthesis pipeline that constructs search tasks spanning both deterministic question answering and open-ended research scenarios with task appropriate evaluation metrics. We train an end-to-end agent using supervised fine-tuning and reinforcement learning, achieving strong and often state of the art performance across benchmarks including BrowseComp (48.6\%), GAIA (75.7\%), Xbench (82.0\%), and DeepResearch Bench (45.9\%). Compared to Mirothinker-v1.0, SMTL with maximum 100 interaction steps reduces the average number of reasoning steps on BrowseComp by 70.7\%, while improving accuracy.