The Year in AI Papers
The year/Independent research

Paper 2511.07327

IterResearch: Rethinking Long-Horizon Agents with Interaction Scaling

Guoxin Chen, Zile Qiao, Xuanzhong Chen, Donglei Yu, Haotian Xu, Wayne Xin Zhao, Ruihua Song, Wenbiao Yin, Huifeng Yin, Liwen Zhang, Kuan Li, Minpeng Liao, et al.

Published
Nov 2025
Research lab
Independent
Citations
14
GitHub
Not linked

01 In brief

Summary

IterResearch is a novel iterative deep-research paradigm that addresses the limitations of mono-contextual approaches, which accumulate all information in a single expanding context window, leading to context suffocation and noise contamination.

It uses an MDP-inspired architecture with strategic workspace reconstruction, maintaining an evolving report as memory and periodically synthesizing insights to preserve reasoning capacity across arbitrary exploration depths.

Training employs Efficiency-Aware Policy Optimization (EAPO), which uses geometric reward discounting to incentivize efficient exploration and adaptive downsampling for stable distributed training.

Experiments show IterResearch outperforms existing open-source agents by an average of +14.5 percentage points across six benchmarks, narrowing the gap with proprietary systems.

It demonstrates unprecedented interaction scaling, extending to 2048 interactions with performance gains from 3.5% to 42.5%, and serves as an effective prompting strategy, improving frontier models by up to 19.2pp over ReAct on long-horizon tasks.

The paradigm also enables cross-paradigm knowledge transfer, enhancing mono-contextual agents when trained on its trajectories.

02 From the paper

Abstract

Recent advances in deep-research agents have shown promise for autonomous knowledge construction through dynamic reasoning over external sources. However, existing approaches rely on a mono-contextual paradigm that accumulates all information in a single, expanding context window, leading to context suffocation and noise contamination that limit their effectiveness on long-horizon tasks. We introduce \textbf{IterResearch}, a novel iterative deep-research paradigm that revisits long-horizon research through the lens of Interaction Scaling. Instead of relying on linear context accumulation, we adopt an MDP-inspired architecture with strategic workspace reconstruction. By maintaining an evolving report as memory and periodically synthesizing insights, our approach preserves consistent reasoning capacity across arbitrary exploration depths. To effectively train this paradigm, we employ Efficiency-Aware Policy Optimization (EAPO), a training strategy that adapts geometric reward discounting to incentivize efficient exploration and utilizes adaptive downsampling for stable distributed training. Extensive experiments demonstrate that IterResearch achieves substantial improvements over existing open-source agents with average +14.5pp across six benchmarks and narrows the gap with frontier proprietary systems. Remarkably, our paradigm exhibits unprecedented interaction scaling, extending to 2048 interactions with dramatic performance gains (from 3.5\% to 42.5\%), and serves as an effective prompting strategy, improving frontier models by up to 19.2pp over ReAct on long-horizon tasks. These findings position IterResearch as a versatile solution for long-horizon reasoning, effective both as a trained agent and as a prompting paradigm for frontier models.