PureDocBench - Real Degraded: leaderboard
The real degraded track of PureDocBench (2026), a source-traceable document-parsing benchmark of 1,475 pages: pages captured by phone from prints, photocopies and screens, or as compressed screenshots. The Overall score averages text accuracy, formula CDM and table TEDS on a 0 to 100 scale; higher is better.
Metric: Overall parsing score (0-100): mean of text accuracy (100 times one minus the text normalized edit distance), formula CDM and table TEDS, 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 September 2028. 58 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3.6 Flash | 75.73 | #81 | |
| 2 | Claude Opus 5 | 75.27 | #26 | |
| 3 | GLM-5.3 Flash | 74.99 | #133 | |
| 4 | Qwen 3.8 27B | 73.59 | #221 | |
| 5 | Gemini 3.1 Pro (Preview) | 71.98 | #99 | |
| 6 | Qwen 3.5 122B A10B | 69.85 | #346 | |
| 7 | WeVisDoc-4B | 69.14 | ||
| 8 | TeleOCR | 69.08 | ||
| 9 | Kimi K2.6 | Non-reasoning | 68.02 | #192 |
| 10 | Logics-Parsing-v2 | 67.64 | ||
| 11 | FD-RL | 67.04 | ||
| 12 | OvisOCR2 | 66.49 | ||
| 13 | Qwen 3.5 27B | 65.92 | #375 | |
| 14 | WeVisDoc-2B | 65.68 | ||
| 15 | Qwen 3.5 9B | 65.45 | #828 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=puredocbench-real-degraded · How It Works · Data refreshed daily, snapshot 2026-11-21.