DocParsingBench - Table: 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 table part: TEDS tree-edit similarity of each matched HTML table.

Metric: Table TEDS (0-1). Source: github.com. Saturation forecast: Rough model projection: around 2026. 17 models tracked. Updated 16 Nov 2026.

Top models

#ModelScoreOverall rank
1OXR-1.00.8004
2Chandra-OCR-20.7769
3TeleOCR0.7677
4MinerU2.5-Pro0.7673
5Dots.mocr0.7518
6Jina-OCR-v10.7503
7PaddleOCR-VL-1.60.7495
8OvisOCR20.742
9Youtu-Parsing0.7248
10Infinity-Parser2-Pro0.7248
11GLM-OCR0.7185
12Qianfan-OCR0.6954
13FireRed-OCR0.693
14Surya OCR 20.6695
15LightOnOCR-2-1B0.6634

Interactive version: theaggregate.ai/benchmark?slug=docparsingbench-table · How It Works · Data refreshed daily, snapshot 2026-11-16.