OmniDocBench

End-to-end document parsing — markdown extraction of text, formulas, tables, and reading order from PDF pages. Source: OmniDocBench repo (v1.6 overall score).

Features measured

Vision

Use cases unlocked

Document & image understanding Chart & diagram analysis

Leaderboard

Higher is better · overall score

1 PaddleOCR-VL-1.6 96.3

2 MinerU2.5-Pro 95.8

3 GLM-OCR 95.2

4 PaddleOCR-VL-1.5 94.9

5 PaddleOCR-VL 94.2

6 Qianfan-OCR 93.9

7 Youtu-Parsing 93.7

8 Ovis2.6-30B-A3B 93.7

Bars scaled from 90 so close scores stay legible — the figure at right is the actual overall score.

Benchmark source ↗

← All benchmarks