DocParsingBench - Formula: leaderboard
Systems turn 1,400 pages of real business documents (finance, legal, research, manufacturing, education; Chinese and English) into Markdown; the README does not state how many pages each system was scored on. This board is the display-formula part: CDM F1 of each matched formula, which compares rendered characters rather than LaTeX strings.
Metric: Display formula CDM F1 (0-1). Source: github.com. Saturation forecast: Rough model projection: around 2026. 17 models tracked. Updated 16 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | OXR-1.0 | 0.8576 | |
| 2 | MinerU2.5-Pro | 0.8206 | |
| 3 | Dots.mocr | 0.8199 | |
| 4 | Chandra-OCR-2 | 0.8182 | |
| 5 | Infinity-Parser2-Pro | 0.7852 | |
| 6 | Jina-OCR-v1 | 0.7808 | |
| 7 | TeleOCR | 0.7619 | |
| 8 | Surya OCR 2 | 0.7619 | |
| 9 | GLM-OCR | 0.7594 | |
| 10 | PaddleOCR-VL-1.6 | 0.7559 | |
| 11 | FireRed-OCR | 0.7477 | |
| 12 | HunyuanOCR 1.5 | 0.7462 | |
| 13 | LightOnOCR-2-1B | 0.738 | |
| 14 | OvisOCR2 | 0.7348 | |
| 15 | Youtu-Parsing | 0.7317 |
Interactive version: theaggregate.ai/benchmark?slug=docparsingbench-formula · How It Works · Data refreshed daily, snapshot 2026-11-16.