WebAnchor: Anchoring Agent Planning to Stabilize Long-Horizon Web Reasoning

2026-01-06 · Xinmiao Yu, Liwen Zhang, Xiaocheng Feng, Yong Jiang, Bing Qin, Pengjun Xie, Jingren Zhou arxiv

Large Language Model(LLM)-based agents have shown strong capabilities in web information seeking, with reinforcement learning (RL) becoming a key optimization paradigm. However, planning remains a bottleneck, as existing methods struggle with long-horizon strategies. Our analysis reveals a critical phenomenon, plan anchor, where the first reasoning step disproportionately impacts downstream behavior in long-horizon web reasoning tasks. Current RL algorithms, fail to account for this by uniformly distributing rewards across the trajectory. To address this, we propose Anchor-GRPO, a two-stage RL framework that decouples planning and execution. In Stage 1, the agent optimizes its first-step planning using fine-grained rubrics derived from self-play experiences and human calibration. In Stage 2, execution is aligned with the initial plan through sparse rewards, ensuring stable and efficient tool usage. We evaluate Anchor-GRPO on four benchmarks: BrowseComp, BrowseComp-Zh, GAIA, and XBench-DeepSearch. Across models from 3B to 30B, Anchor-GRPO outperforms baseline GRPO and First-step GRPO, improving task success and tool efficiency. Notably, WebAnchor-30B achieves 46.0% pass@1 on BrowseComp and 76.4% on GAIA. Anchor-GRPO also demonstrates strong scalability, getting higher accuracy as model size and context length increase.

📄 PDF Abstract BibTeX arXiv:2601.03164

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

FORGE: Research-Trajectory Hijacking Attacks on Deep Research Agents

2026-07-06 · Yue Pan, Ziheng Zhang, Junxiang Lei, Changhao Jia 외 arxiv

Deep research agents decompose open-ended queries into subtasks, retrieve web evidence over multiple rounds, and synthesize long-form reports. This workflow creates a planning-layer poisoning surface: adversarial documen…

Continual GUI Agents

2026-01-28 · Ziwei Liu, Borui Kang, Hangjie Yuan, Zixiang Zhao 외 arxiv

As digital environments (data distribution) are in flux, with new GUI data arriving over time-introducing new domains or resolutions-agents trained on static environments deteriorate in performance. In this work, we intr…

Continual Learning

Not All Agents Matter: From Global Attention Dilution to Risk-Prioritized Game Planning

2026-04-07 · Kang Ding, Hongsong Wang, Jie Gui, Lei He arxiv

End-to-end autonomous driving resides not in the integration of perception and planning, but rather in the dynamic multi-agent game within a unified representation space. Most existing end-to-end models treat all agents …

Autonomous Driving Decision Making

Momentum-Anchored Multi-Scale Fusion Model for Long-Tailed Chest X-Ray Classification

2026-05-04 · Duy Hoang Khuong, Duy Nguyen Huu, Ngu Huynh Cong Viet arxiv

Chest X-ray classification suffers from severe class imbalance where gradient updates bias toward majority classes, causing feature drift and poor performance on rare but critical pathologies. We propose a Momentum-Ancho…

Medical Image Classification

HIPIF: Hierarchical Planning and Information Folding for Long-Horizon LLM Agent Learning

2026-06-09 · Juncheng Diao, Zhicong Lu, Peiguang Li, Yongwei Zhou 외 arxiv

While Large Language Models (LLMs) have demonstrated strong capabilities as autonomous agents across a wide range of tasks, their performance often degrades in multi-turn long-horizon agentic tasks. Existing methods have…

Hierarchical Reinforcement Learning