MiniCPM-OCR-1.0: Turn scans and PDFs into editable text (1.3B)

What it does

MiniCPM-OCR-1.0 is a small 1.3B vision-language model that reads text from images. For whole documents, its software kit (SDK) first finds the layout with a separate layout model, PP-DocLayoutV3, then recognizes each region and assembles Markdown, a simple text format with headings and tables, in reading order. It handles text, tables, formulas, charts, seals and chosen fields.

How to use it

Open a web demo, such as the Hugging Face Space or ModelBest's playground, and try a document. For local use, install the project's SDK and pick a backend (the engine that runs the model) such as vLLM, SGLang, llama.cpp or Ollama. A coding agent can also follow the project's skill guide to set up, download the model and parse a first document.

Common use cases

• Upload a photo of a curved or skewed document page, and you get Markdown text in reading order, with the page layout detected.
• Give it a table or chart image, and you get an HTML table; give it a formula image, and you get LaTeX, the code math papers use.
• Send a form image and list the fields you want, and you get a JSON data object with those fields, and ### marking any not on the page.