AREX: Towards a Recursively Self-Improving Agent for Deep Research

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Shuqi Lu, Chaofan Li, Kun Luo, Zhang Zhang, Hui Wang, Hongwang Xiao, Zheng Liu, Lei Xiong, Jiahao Wang, Sen Wang, Xiyan Jiang, Wanli Li, Yuyang Hu, Hongjin Qian, Bingyu Yan, Ziyi Xia, Yingxia Shao, Kang Liu, Zhicheng Dou, Di He, Chaozhuo Li, Qiwei Ye, Zhongyuan Wang, Zheng Liu
• Venue: arXiv: cs.AI
• Published: 2026-07-23
• Collected: 2026-07-24
• URL: http://arxiv.org/abs/2607.21461v1
• PDF: https://arxiv.org/pdf/2607.21461v1
• DOI: UNKNOWN

Plain-Language Brief

The paper introduces a new type of deep research agent called AREX. These agents are designed to recursively improve their answers by verifying intermediate results, using the partially verified state to guide subsequent refinement. The agents learn an autonomous context-update tool that compresses growing interaction history into a compact improvement state. The authors trained AREX on synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning. The results show that AREX outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.

What the Document Says

• The paper presents a family of Recursively Self-Improving (RSI) deep research agents called AREX.
• AREX alternates between an inner research loop and an outer self-improvement loop.
• The inner research loop gathers evidence and constructs a provisional answer.
• The outer self-improvement loop audits the answer, identifies unresolved claims, and launches targeted follow-up research.
• AREX learns an autonomous context-update tool for sustaining RSI over long horizons.
• Training of AREX is done on verified synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning.
• The paper emphasizes key steps where decisive evidence is acquired or erroneous research directions are corrected to mitigate sparse final rewards during long horizon learning.
• The authors instantiate a dense 4B model and a 122B-A10B Mixture-of-Experts model of AREX.
• Across various reasoning and tool-use benchmarks, including BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), AREX outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.

Key Concepts Mentioned

• Deep research
• Recursively Self-Improving (RSI) agents
• AREX
• inner research loop
• outer self-improvement loop
• context-update tool
• agentic mid-training
• long-horizon reinforcement learning
• decisive evidence
• erroneous research directions
• synthetic tasks
• high-quality trajectories
• dense 4B model
• 122B-A10B Mixture-of-Experts model
• benchmarks

Claims Made by the Document

• AREX outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.

Evidence Used

• Across various reasoning and tool-use benchmarks, including BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), AREX outperforms comparable-scale baselines.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery--verification asymmetry suggests that a research agent should do more than simply search longer: it should recursively improve its current answer by verifying intermediate results and using the partially verified state to guide subsequent refinement. We introduce AREX, a family of Recursively Self-Improving (RSI) deep research agents. AREX alternates between an inner research loop that gathers evidence and constructs a provisional answer, and an outer self-improvement loop that audits the answer constraint-wise, identifies unresolved claims, and launches targeted follow-up research. To sustain RSI over long horizons, AREX learns an autonomous context-update tool that compresses growing interaction history into a compact improvement state preserving verified evidence and unresolved constraints, without relying on an external model. We train AREX on verified synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning. To mitigate sparse final rewards during long horizon learning, we emphasize key steps where decisive evidence is acquired or erroneous research directions are corrected. We instantiate a dense 4B model and a 122B-A10B Mixture-of-Experts model. Across BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), and other reasoning and tool-use benchmarks, AREX substantially outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.