OPEN LINK ↗
// 1h agoMODEL RELEASE
MiniCPM-OCR Pairs 1.3B Parser With Jev
OpenBMB released MiniCPM-OCR 1.0, a 1.3B open-weight model for extracting text, tables, and formulas from difficult documents, alongside MiniCPM-OCR Jev, a separate model that scores parsing reliability. The two-model pipeline is designed to automate easy cases while routing uncertain results to human reviewers. [Official model page](https://openbmb.cn/en/model/minicpm-ocr)
// ANALYSIS
Jev is the more consequential idea here: OCR systems need quality control, not just better transcription, especially when silent table and number errors can contaminate downstream RAG systems.
- –Jev evaluates text, formulas, and tables against the source image and can judge outputs from other OCR engines.
- –Its block-level confidence scores enable selective human review instead of manually checking every page.
- –MiniCPM-OCR supports local deployment through vLLM, SGLang, llama.cpp, Ollama, and quantized checkpoints.
- –OpenBMB reports strong benchmark results and 55% higher efficiency with DSpark, but independent validation is still needed before trusting its confidence scores in production.
- –The Apache-2.0 release makes this practical for teams building private document-ingestion pipelines. [GitHub repository](https://github.com/OpenBMB/MiniCPM-OCR)
// TAGS
minicpm-ocrllmsmall-llmopen-weightsmultimodalvisionocropen-source
DISCOVERED
1h ago
2026-10-28
PUBLISHED
1h ago
2026-10-28
RELEVANCE
9/ 10
AUTHOR
dani_avila7