Turning pages into structured text. Now dominated by vision language models rather than the classical OCR engines. · ranked by document parsing benchmarks · verified 2 Sept 2026

#	Name	OmniDocBench	Open weights	Handles	Licence	Kind	Status
1	GLM-OCR Zhipu AI	94.6	Yes	Tables, formulas, handwriting	Open weights	Vision language model	current
2	dots.ocr Xiaohongshu	~93	Yes	Layout, 100+ languages	MIT	Vision language model	current
3	Qwen3-VL Alibaba	~93	Yes	Documents, charts, video	Apache 2.0	Vision language model	current
4	DeepSeek-OCR DeepSeek	~92	Yes	Dense text, tables	MIT	Vision language model	current
5	Mistral OCR Mistral AI	~92	No	Tables, images, equations	Proprietary	Hosted API	current
6	CCohere Parse 5 Cohere	79.2 on ParseBench	No	Tables as HTML, forms, diagrams, bounding boxes	Proprietary	Vision language model, 2.3B	current
7	olmOCR Allen Institute for AI	~91	Yes	Tables, markdown structure	Apache 2.0	Vision language model	current
8	GOT-OCR 2.0 StepFun	~90	Yes	Formulas, music, charts	Apache 2.0	Vision language model	current
9	MinerU OpenDataLab	~90	Yes	PDF, formulas, tables	AGPL-3.0	Pipeline	current
10	Docling IBM	~88	Yes	PDF, DOCX, PPTX, HTML	MIT	Pipeline	current
11	Surya Datalab	~87	Yes	Layout, reading order, tables	GPL-3.0	Pipeline	current
12	PaddleOCR Baidu	~82	Yes	Text, tables, receipts	Apache 2.0	Classical engine	current
13	Tesseract Google	~55	Yes	Clean printed text	Apache 2.0	Classical engine	current

#1

GLM-OCR

Zhipu AI

94.6

OmniDocBench v1.5

• Ahead of Gemini 3 Pro and GPT-5.2 on the same benchmark
• 94.0 on OCRBench

Best forcomplex documents end to end

#2

dots.ocr

Xiaohongshu

1.7B

parameters

• Layout and content in one pass
• Strong for its size

Best formultilingual layout parsing