# AmbiHTMLBench: A Benchmark for HTML-based Visually Rich Document parsing

DevFeed: [AmbiHTMLBench: A Benchmark for HTML-based Visually Rich Document parsing](<https://devfeed.tech/articles/ambihtmlbench-a-benchmark-for-html-based-visually-rich-document-parsing-118469.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.06449>)

Author: Boyu Niu, Yingjie Xiong, Shurui Li, Linhan Pang, Yuntong Zhang, Han Wang, Hao Wu, Kai Zuo, Xu Tang, Xuanhe Zhou, Fan Wu

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [HTML](<https://devfeed.tech/topics/html.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Markdown](<https://devfeed.tech/topics/markdown.md>)

Tags: [annotations](<https://devfeed.tech/tags/annotations.md>), [appearance](<https://devfeed.tech/tags/appearance.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [error](<https://devfeed.tech/tags/error.md>), [html](<https://devfeed.tech/tags/html.md>), [layout](<https://devfeed.tech/tags/layout.md>), [markdown](<https://devfeed.tech/tags/markdown.md>), [model](<https://devfeed.tech/tags/model.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [organization](<https://devfeed.tech/tags/organization.md>), [pages](<https://devfeed.tech/tags/pages.md>), [parsing](<https://devfeed.tech/tags/parsing.md>), [rendering](<https://devfeed.tech/tags/rendering.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [structure](<https://devfeed.tech/tags/structure.md>), [table](<https://devfeed.tech/tags/table.md>), [testbed](<https://devfeed.tech/tags/testbed.md>), [text](<https://devfeed.tech/tags/text.md>), [visual-hierarchy](<https://devfeed.tech/tags/visual-hierarchy.md>)

## AI overview

AmbiHTMLBench evaluates whether models can reconstruct real-world documents as executable HTML that preserves content, organization, and appearance. It includes 1,115 annotated pages and compares systems using rendered-block alignment for table fidelity, page-level text error, and two-dimensional layout consistency, alongside a reference-anchored assessment that aligns LLM judges with human preferences. Among ten primary multimodal systems, the top objective score was 0.8092, but a different system won the subjective evaluation, suggesting the protocols capture complementary aspects of fidelity. The authors also report that task-specific fine-tuning of a small-parameter expert model performs well on document-to-HTML reconstruction.

## Source excerpt

arXiv:2611.06449v1 Announce Type: new Abstract: Traditional document parsing systems use Markdown as their output representation, which is effective for linear text but does not fully preserve fine-grained spatial structure, table geometry, visual hierarchy, or rendering semantics. We show that HTML is advantageous for heterogeneous and layout-sensitive documents whose representations must remain editable, inspectable, and reusable, and introduce AmbiHTMLBench to systematically study this setting. In this benchmark, given a real-world document, a model must recover executable HTML that preserves its content, organization, and visual appearance. The benchmark contains 1,115 diverse pages with structured annotations and evaluates predictions in two different ways: The objective protocol aligns ground-truth and predicted rendered blocks and measures table fidelity, page-level text error, and a two-dimensional layout consistency; a complementary reference-anchored protocol measures perceived fidelity by aligning LLM judges with human preferences. Across ten primary multimodal systems, the best objective score is 0.8092, while the subjective winner is different, demonstrating that automatic decomposition and human-perceived fidelity capture complementary properties. We further fine-tune a small-parameter expert model and find that task-specific adaptation yields good performance on document-to-HTML reconstruction task. These results position AmbiHTMLBench as a diagnostic testbed for building smarter and more logical document parsing systems.