OpenBMB releases MiniCPM-OCR 1.0 and MiniCPM-OCR-Jev for local document parsing and verification

·2026.10.29 01:31

Key point

The 1B OCR model runs locally on mobile devices while the 2B verifier model checks output accuracy against original images.

OpenBMB released MiniCPM-OCR 1.0 (1B) and MiniCPM-OCR-Jev (2B), both with open weights.
MiniCPM-OCR 1.0 runs locally on iPhone and iPad, parsing handwritten notes, formulas, and tables without cloud dependency.
MiniCPM-OCR 1.0 achieves a score of 93.68 on Real-5 OmniDocBench and 54.7 on OCRBench v2.
MiniCPM-OCR-Jev evaluates parsed text, formulas, and tables against the original image, assigning correctness scores from 0 to 1.
The verifier approaches Qwen3.5-397B-A17B performance on JevOCRBench and works with outputs from different OCR parsers.
Both models support tools including llama.cpp, vLLM, SGLang, and Ollama.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.