PureDocBench - Clean - Reading Order Edit Distance: leaderboard
Metric: Reading-order normalized edit distance (%) between the predicted and reference block sequences, times 100, on the clean track (pages rendered straight from their HTML/CSS source) 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.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.8 27B | 30.5 |
| 2 | Claude Opus 5 | 33.6 |
| 3 | Qwen 3.5 27B | 36.2 |
| 4 | Qwen 3.5 397B A17B | 36.6 |
| 5 | Qwen 3.5 35B A3B | 37.4 |
| 6 | Qwen 3.5 122B A10B | 37.5 |
| 7 | GLM-5.3 Flash | 37.5 |
| 8 | Gemini 3.6 Flash | 38.1 |
| 9 | Qwen 3.5 9B | 38.8 |
| 10 | Qwen 3.6 35B A3B | 39.7 |
| 11 | Gemini 3.1 Pro (Preview) | 40.9 |
| 12 | Qwen 3.5 4B | 41 |
| 13 | Qwen3.8-Flash-Next | 42 |
| 14 | GPT-5.6 Sol | 44 |
| 15 | Qwen 3.6 27B | 44.6 |
Interactive version: theaggregate.ai/benchmark?slug=puredocbench-clean-reading-order-edit-distance · How It Works · Data refreshed daily, snapshot 2026-10-07.