InduOCRBench - Reading Order: leaderboard
The reading-order score of InduOCRBench (2026), a document-parsing benchmark of 570 real industrial documents in 11 challenge types, including historical books with non-standard reading orders: one minus the normalized edit distance of the block order, times 100. Higher is better.
Metric: Reading-order edit-distance similarity (0-100): one minus the normalized edit distance between the predicted and the reference block order, times 100, on 570 industrial documents (3,402 pages) in 11 challenge types (complex background, high pixel count, ultra-long, multi-column, ultra-wide, historical books, handwriting, multi-font, visual style, watermark, cross-page tables) plus normal documents, each OCR method converting pages to structured Markdown (general VLMs with a standardized prompt, pipeline tools at default configuration, closed solutions through their official APIs); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | HunyuanOCR | 85.7 | |
| 2 | PaddleOCR-VL | 85.6 | |
| 3 | MinerU2.5-2509-1.2B | 84.4 | |
| 4 | Qwen 3 VL 235B A22B Instruct | 82.1 | #532 |
| 5 | DeepSeek-OCR | 81.8 | |
| 6 | Mineru2-pipeline | 81.3 | |
| 7 | Doc2X | 81.3 | |
| 8 | Gemini 2.5 Pro | 81.1 | #276 |
| 9 | PP-StructureV3 | 79.1 | |
| 10 | GPT-4o | 70 | #751 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=induocrbench-reading-order · How It Works · Data refreshed daily, snapshot 2026-11-21.