Nanonets-OCR-s
Nanonets-OCR-s is Nanonets' first open image-to-Markdown OCR model, built for documents that a language model will read next.
A release of Nanonets-OCR. Other versions: Nanonets-OCR2.
- Released
- June 12, 2025
- Model ID
- nanonets/Nanonets-OCR-s
- Size
- About 4B, fine-tuned from Qwen2.5-VL-3B-Instruct
- License
- See the Hugging Face model card
- Also searched as
- nanonets-ocr-s, nanonets ocr s, nanonets/Nanonets-OCR-s
What is new in Nanonets-OCR-s
- Converts equations to LaTeX and tells inline math from display math.
- Describes images and charts inside <img> tags.
- Puts signatures and watermark text in their own tags.
- Turns checkboxes and radio buttons into ☐, ☑ and ☒.
- Extracts complex tables as Markdown and HTML. Trained on over 250,000 pages of papers, financial, legal and healthcare documents, tax forms, receipts and invoices.
Use Nanonets-OCR-s through an API
OCRRouter runs a different release of Nanonets-OCR as nanonets/nanonets-ocr.
Nanonets-OCR versions
- Nanonets-OCR: all versions, price, speed and features on OCRRouter
- Nanonets-OCR2 (October 2025)
- Nanonets-OCR-s (June 12, 2025), this page
Questions
Sources
Last updated .