Picture: Illustration · BitBytec · Original illustration, BitBytec
ModelBest and OpenBMB jointly announced the release and open-source availability of the MiniCPM-OCR series on 31 October 2026. This family of models is designed for efficient document parsing and parsing-quality assessment.
Key facts * The launch includes two specific models: MiniCPM-OCR 1.0 (1.3B) and MiniCPM-OCR Jev (2B) [1]. * MiniCPM-OCR 1.0 is the first OCR-focused model in the MiniCPM family, featuring a compact architecture optimised for edge deployment [1]. * MiniCPM-OCR Jev functions as an "AI quality inspector" that assigns confidence scores to text, LaTeX formulas, and tables [1]. * The series supports multiple frameworks including vLLM, SGLang, llama.cpp, and Ollama [1].
What do the MiniCPM-OCR models do? MiniCPM-OCR 1.0 provides high-accuracy parsing for scanned, photographed, curved, tilted, and screen-captured documents without requiring image pre-correction [1]. It is capable of handling dense bilingual Chinese-English content, complex mathematical formulas, and multi-level or borderless tables [1].
Beyond full-page reconstruction, MiniCPM-OCR 1.0 supports key information extraction (KIE) to pull specific fields from invoices, IDs, and financial statements based on business instructions [1]. The model uses DSpark decoding acceleration to increase end-to-end throughput and lower per-page latency [1].
How does MiniCPM-OCR Jev work? MiniCPM-OCR Jev acts as a decision model that compares a source image with OCR output to detect missing characters, errors, and formatting misalignments [1]. It identifies reliable results for direct processing and flags inaccurate outputs for human review [1].
Who can use the MiniCPM-OCR series? Enterprises processing large volumes of documents, such as financial contract archives, academic libraries, and government records, can use the models to reduce computing costs [1]. The series is available on GitHub, Hugging Face, and ModelScope, with Day-0 hardware adaptation for AMD, Arm, Houmo AI, Intel, and NVIDIA [1]. It also features Day-0 support for various edge chips through the FlagOS software stack [1].
Sources
- ModelBest and OpenBMB Open-Source MiniCPM-OCR Series, Delivering High-Performance Document Parsing and AI-Powered Quality Control — Cision PR Newswire (primary source)
- The Epstein Files: How the saga unfolded within Trump’s administration | The Independent — The Independent
- Jeffrey Epstein list: Who is named in court filings? — bbc.com
Written with AI from the sources listed, checked against them before publishing.