# Let's talk document formatting.

Bold. Italics. Superscripts. Strikethroughs. The visual cues humans...

Canonical URL: https://www.traeai.com/articles/6f726fa8-a986-48fa-b5ce-be4b2d62c3cf
Original source: https://x.com/llama_index/status/2049139409316946011
Source name: LlamaIndex 🦙(@llama_index)
Content type: tweet
Language: 中文
Score: 7.8
Reading time: 1 分钟
Published: 2026-04-28T14:51:21+00:00
Tags: OCR, AI Agent, Document Understanding, LlamaIndex, Benchmark

## Summary

LlamaIndex 发布 ParseBench——首个面向 AI 代理的文档 OCR 基准，首次将语义格式（加粗/删除线/上标等）纳入评估，强调视觉格式即语义。

## Key Takeaways

- 现有 OCR 基准完全忽略加粗、删除线、上标等人类依赖的语义格式线索
- ‘$199’删除线+‘$149’并列不是装饰，而是价格对比的核心语义
- ParseBench 引入 Semantic Formatting Score，是首个专为 AI 代理理解文档设计的 OCR 基准

## Outline

- 问题提出 — 指出当前 OCR 基准忽视文档视觉格式所承载的关键语义信息。
  - 语义格式即意义 — 以删除线价格、上标引用为例，说明格式改变文本真实含义。
  - ParseBench 发布 — 介绍新基准的定位：首个面向 AI 代理的文档 OCR 评测框架。
    - Semantic Formatting Score — 五大指标之一，量化模型保留原始文档格式语义的能力。

## Highlights

- > Bold. Italics. Superscripts. Strikethroughs. The visual cues humans rely on every time we read a doc, and ones existing OCR benchmarks completely ignore. — 第 1 段
- > “$199” struck through next to “$149” isn't decoration. It's the meaning. — 第 2 段
- > A superscript tells your agent "3" is a citation, not part of the number. Flatten that and your agent is reading a different doc than you are. — 第 3 段
- > Two weeks ago we released ParseBench, the first document OCR benchmark for AI agents. — 第 4 段

## Citation Guidance

When citing this item, prefer the canonical traeai article URL for the AI-readable summary and include the original source URL when discussing the underlying source material.