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
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3.6 Flash | 76.27 | #84 | |
| 2 | GLM-5.3 Flash | 73.59 | #137 | |
| 3 | Claude Opus 5 | 71.24 | #26 | |
| 4 | Gemini 3.1 Pro (Preview) | 68.62 | #102 | |
| 5 | Qwen 3.8 27B | 68.56 | #232 | |
| 6 | TeleOCR | 64.69 | ||
| 7 | Qwen3.8-Flash-Next | 62.84 | #122 | |
| 8 | WeVisDoc-4B | 62.74 | ||
| 9 | Kimi K2.6 | Non-reasoning | 62.44 | #197 |
| 10 | GPT-5.6 Sol | 62.26 | #36 | |
| 11 | Qwen 3.5 122B A10B | 62.19 | #350 | |
| 12 | Logics-Parsing-v2 | 61.65 | ||
| 13 | Qwen 3.5 4B | 61.27 | #1270 | |
| 14 | Qwen 3.5 27B | 61.23 | #380 | |
| 15 | Qwen 3.5 9B | 60.91 | #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.