DocAtlas: Long-Document Understanding as Mutable-State Interaction

Abstract: Long-document understanding requires models to find and combine evidence across many pages, layouts, tables, figures, and charts. Existing retrieval-augmented systems usually select evidence from a static index before generation, while recent agentic systems add multi-turn tool use but often rely on frozen proprietary backbones whose behavior is set by prompts.

摘要： 长文档理解要求模型能够跨越多个页面、布局、表格、图表和图形来查找并整合证据。现有的检索增强系统通常在生成前从静态索引中选择证据，而近期的智能体系统虽然增加了多轮工具使用，但往往依赖于行为由提示词（prompts）设定的冻结专有模型骨干。

We present DocAtlas, a system that treats long-document understanding as a mutable-state information-seeking process. We instantiate DocAtlas as a mutable document harness: an external environment that determines what document information is searched, read, stored, reviewed, and shown to the model at each step.

我们提出了 DocAtlas，这是一个将长文档理解视为“可变状态信息检索过程”的系统。我们将 DocAtlas 实例化为一个可变文档工具框架（mutable document harness）：这是一个外部环境，它决定了在每一步中搜索、阅读、存储、回顾并展示给模型的文档信息。

Given a document and question, the harness exposes search, reading, note-taking, and review tools, maintains a hierarchical tree and note store, and updates both as the agent records evidence. DocAtlas combines self-improving retrieval, selective evidence access, and active working memory under a fixed context budget.

给定文档和问题后，该框架会提供搜索、阅读、笔记和回顾工具，维护一个层级树和笔记存储库，并在智能体记录证据时同步更新两者。DocAtlas 在固定的上下文预算下，结合了自我改进的检索、选择性证据访问和主动工作记忆。

The same harness supports inference-time use with large VLMs and end-to-end reinforcement learning for compact VLM agents. With GPT-5.4, DocAtlas reaches 71.4% on MMLongBench-Doc, exceeding the human-expert reference of 65.8%. A Qwen3.5-4B VLM trained with end-to-end RL in the DocAtlas environment reaches 63.7%, compared with a 54.4% direct-input baseline, showing that mutable document-harness design can improve compact document agents by a large margin.

同一框架既支持大型视觉语言模型（VLM）在推理阶段的使用，也支持小型 VLM 智能体的端到端强化学习。在使用 GPT-5.4 的情况下，DocAtlas 在 MMLongBench-Doc 基准测试中达到了 71.4% 的准确率，超过了 65.8% 的人类专家参考水平。而在 DocAtlas 环境中通过端到端强化学习训练的 Qwen3.5-4B VLM 达到了 63.7% 的准确率，相比之下直接输入基准仅为 54.4%，这表明可变文档框架设计可以显著提升小型文档智能体的性能。