openbmb/minicpm-ocr is an OCR repository configured to read the contents of documents and PDFs and to review recognition results. The repository publishes the MiniCPM-OCR 1.0 and MiniCPM-OCR Jev models. MiniCPM-OCR 1.0 is a 1.3B model dedicated to document reading, and Jev is a 2B verification model designed to find per-area review points within recognition results. No separate recent release is recorded. The repository description and the content presented in the README are the basis for the public scope.
MiniCPM-OCR 1.0 reads not only characters but also structured content such as tables, formulas, and charts in documents, and Jev finds the areas worth revisiting in those results. In other words, it is a setup that divides document recognition and result review within a single model family. The repository uses Python and follows the Apache-2.0 license. The README claims support for llama.cpp, vLLM, sglang, and ollama, and provides inference and deployment guidance along with reproducible performance evaluation documents. This shows the direction of being able to configure OCR across multiple inference environments, but it does not mean that the behavior in each environment was independently verified.
The performance comparison in the README is a result presented by the project, so it cannot be treated as an independently verified value. There is also nothing confirmed beyond the README's description regarding the model architecture and actual performance. Therefore, what this repository shows is a design that handles document recognition and result review together, along with deployment guidance, and the accuracy or performance needed for real work must be checked separately.