MiniCPM-OCR series
ModelBest and OpenBMB open-sourced the MiniCPM-OCR series, with a 1.3B document parsing model and a 2B model that scores OCR output quality.
MiniCPM-OCR 1.0 (1.3B) is the first OCR-focused model in the MiniCPM family. It parses scanned, photographed, curved, tilted and screen-captured documents without image pre-correction, handles Chinese-English text, formulas and borderless tables, and does key information extraction. MiniCPM-OCR Jev (2B) does not run OCR. It compares the source image with OCR output and assigns confidence scores to text, LaTeX and tables, so it can flag results for human review or pick the best of several OCR systems.
- Date
- Saturday 31 October 2026
- Lab
- ModelBest
- Kind
- open-weights
- Access
- open weights
Paid press release. Available on GitHub, Hugging Face and ModelScope with vLLM, SGLang, llama.cpp and Ollama support. License and benchmark figures are not stated in the release.
Sources
This record was checked against its sources on 1 November 2026. How we check