Back to News
ATH-MaaS (Alibaba Token Hub Model-as-a-Service)•July 15, 2026
OvisOCR2
Open Source
ocr
document-parsing
multimodal
markdown
### TL;DR
OvisOCR2 is a compact 0.8B parameter end-to-end document-parsing model that transforms document page images into structured Markdown. By eliminating the need for complex multi-stage pipelines (layout detection, OCR, and reassembly), it provides a streamlined solution for extracting text, tables, formulas, and visual regions in natural reading order.
Key Insights & Metrics
Pricing
Free & Open source
Cost structure
Version
2
Current release version
Hardware
An entry-level GPU with 4 GB to 8 GB of VRAM and roughly 3 GB of free disk space to run this highly compact model locally
Compute requirements
Category
Open Source
Licensing model
Region
China
Primary region
Key Features
- End-to-end document parsing in a single pass
- Outputs structured Markdown with HTML tables and LaTeX formulas
- Highly efficient 0.8B parameter size suitable for local and edge deployment
- State-of-the-art performance (96.58 on OmniDocBench v1.6)
- Full compatibility with vLLM, Transformers, and SGLang inference stacks
Discussion
0
Upvotes
0
Downvotes
0 reviews
Sign in to leave a review
Reviews
No reviews yet. Be the first to review!