PureDocBench - Real Degraded - Text Edit Distance: leaderboard

Text normalized edit distance on the real degraded track of PureDocBench (2026), a document-parsing benchmark of 1,475 source-traceable pages (pages captured by phone from prints, photocopies and screens, or as compressed screenshots). Percent; lower is better.

Metric: Text normalized edit distance (%): edit distance between the predicted and reference text divided by the longer length, times 100, on the real degraded track (the same pages through four physical acquisition chains: phone photos of prints and of photocopies, screen photographs, and screenshots with social-media compression) of 1,475 source pages in 10 domains and 66 subcategories, rendered from LLM-written HTML/CSS with source-linked annotations for text, formulas, tables and reading order; lower is better. Source: arxiv.org. Saturation forecast: Not forecast. 58 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.8 27B27.4#232
2Qwen 3.5 122B A10B28.1#350
3Qwen 3.5 27B28.3#380
4Qwen 3.5 397B A17B28.7#270
5Claude Opus 529.4#26
6FD-RL29.8
7Gemini 3.1 Pro (Preview)30#102
8Logics-Parsing-v230.4
9Qwen 3.5 35B A3B31#494
10Gemini 3.6 Flash31#84
11GLM-5.3 Flash31.1#137
12Dots.mocr31.2
13OvisOCR232
14Qwen 3.5 9B33.2#838
15WeVisDoc-4B33.3

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=puredocbench-real-degraded-text-edit-distance · How It Works · Data refreshed daily, snapshot 2026-11-22.