The Year in AI Papers
The year/Independent research

Paper 2607.21461

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

Shuqi Lu, Chaofan Li, Kun Luo, Zhang Zhang, Hui Wang, Hongwang Xiao, Lei Xiong, Jiahao Wang, Sen Wang, Xiyan Jiang, Wanli Li, Yuyang Hu, et al.

Published
Jul 2026
Research lab
Independent
Citations
0
GitHub
Not linked

01 In brief

Summary

AREX is a family of recursively self-improving deep research agents developed by BAAI.

It exploits the discovery-verification asymmetry in deep research by alternating between an inner research loop that gathers evidence and constructs provisional answers, and an outer self-improvement loop that audits answers constraint-wise, identifies unresolved claims, and launches targeted follow-up research.

AREX learns an autonomous context-update tool to compress interaction history into a compact improvement state, preserving verified evidence and unresolved constraints.

Training involves verified synthetic tasks, agentic mid-training, and long-horizon reinforcement learning with key-step emphasis.

Instantiations include a dense 4B model (Turbo) and a 122B-A10B MoE model (Base).

Across benchmarks like BrowseComp, WideSearch, DeepSearchQA, and HLE, AREX outperforms comparable-scale baselines and remains competitive with larger models.

Ablations show that progressive capability training, key-step focused supervision, and step-aware RL each contribute to performance, with full AREX achieving 82.5% on BrowseComp versus 77.5% without progressive training, 74.1% without key-step supervision, and 79.4% without step-aware RL.

The outer loop and context updating together improve accuracy by 22.9 points on BrowseComp.

Future work will explore more general step-utility estimation mechanisms.

02 From the paper

Abstract

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.