PureDocBench - Real Degraded - Formula CDM: leaderboard
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 2029. 58 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.6 Flash | 76.27 |
| 2 | GLM-5.3 Flash | 73.59 |
| 3 | Claude Opus 5 | 71.24 |
| 4 | Gemini 3.1 Pro (Preview) | 68.62 |
| 5 | Qwen 3.8 27B | 68.56 |
| 6 | Qwen3.8-Flash-Next | 62.84 |
| 7 | Kimi K2.6 (Non-reasoning) | 62.44 |
| 8 | GPT-5.6 Sol | 62.26 |
| 9 | Qwen 3.5 122B A10B | 62.19 |
| 10 | Qwen 3.5 4B | 61.27 |
| 11 | Qwen 3.5 27B | 61.23 |
| 12 | Qwen 3.5 9B | 60.91 |
| 13 | Qwen 3.5 397B A17B | 60.7 |
| 14 | Qwen 3.5 35B A3B | 59.68 |
| 15 | Qwen 3.6 35B A3B | 59.66 |
Interactive version: theaggregate.ai/benchmark?slug=puredocbench-real-degraded-formula-cdm · How It Works · Data refreshed daily, snapshot 2026-10-07.