PureDocBench - Real Degraded - Formula CDM: leaderboard

Formula recognition 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), scored by Character Detection Matching of rendered formulas from 0 to 100. Higher is better.

Metric: Formula Character Detection Matching score (0-100) between the rendered predicted and reference LaTeX formulas, 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; higher is better. Source: arxiv.org. Saturation forecast: Around October 2028. 58 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.6 Flash76.27#84
2GLM-5.3 Flash73.59#137
3Claude Opus 571.24#26
4Gemini 3.1 Pro (Preview)68.62#102
5Qwen 3.8 27B68.56#232
6TeleOCR64.69
7Qwen3.8-Flash-Next62.84#122
8WeVisDoc-4B62.74
9Kimi K2.6Non-reasoning62.44#197
10GPT-5.6 Sol62.26#36
11Qwen 3.5 122B A10B62.19#350
12Logics-Parsing-v261.65
13Qwen 3.5 4B61.27#1270
14Qwen 3.5 27B61.23#380
15Qwen 3.5 9B60.91#838

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-formula-cdm · How It Works · Data refreshed daily, snapshot 2026-11-22.