MiniCPM OCR

MiniCPM OCR positions itself as a document parsing workspace rather than a plain converter. Instead of returning a wall of text, it keeps the original page visible while presenting recognized content in a structured, editable form. The site is built around the MiniCPM-OCR 1.0 model and targets documents that mix prose, tables, and mathematical notation, such as research notes, financial statements, and scanned books.

The core workflow is upload, review, edit, and export. You can drop in a PDF, PNG, JPEG, GIF, WebP, or BMP file, with limits of 24 MiB for a single image and 64 MiB or 500 pages for a PDF. Recognition output is linked to regions on the source page, so clicking a block shows where it came from. Results can be corrected in place and then exported as editable Word, Markdown with tables and LaTeX, or JSON carrying region and coordinate data.

Typical users include researchers pulling formulas out of papers, analysts digitizing financial tables, students converting scanned textbook pages, and developers who need structured JSON for a pipeline. Anyone retyping content from screenshots or image-only PDFs is a natural fit, especially when layout and formulas must survive the conversion.

Start with the free single-document trial of up to three pages and no sign-up to judge accuracy on your own material. Because billing runs at one credit per page, estimate page counts before committing to a plan, and always spot-check tables and complex formulas against the original. Th