ModelBest and OpenBMB Release MiniCPM-OCR 1.0 and Jev for Document Parsing and Quality Control

ModelBest and OpenBMB released and open-sourced the MiniCPM-OCR series on October 31, 2026, comprising MiniCPM-OCR 1.0 (1.3B) and MiniCPM-OCR Jev (2B) for document parsing and parsing-quality assessment. PR Newswire carried the announcement from Beijing.

MiniCPM-OCR 1.0 and Jev Ship as a Pair

MiniCPM-OCR 1.0 is the first OCR-focused model in the MiniCPM family, built on a 1.3-billion-parameter architecture optimized for edge deployment. The model parses scanned, photographed, curved, tilted and screen-captured documents without image pre-correction, and handles dense bilingual Chinese-English content, complex mathematical formulas and multi-level or borderless tables. It also supports key information extraction, pulling specified fields from invoices, IDs and financial statements according to business instructions. DSpark decoding acceleration raises end-to-end throughput and lowers per-page latency, which the company ties to lower computing cost for enterprises processing millions of pages.

MiniCPM-OCR Jev is described as the first OCR result decision model, acting as an AI quality inspector. Jev does not perform OCR itself; it compares the source image with OCR output and assigns confidence scores to text, LaTeX formulas and tables, detecting missing characters, errors and formatting misalignments. The 2B-parameter model supports cross-model deployment and can be integrated into existing OCR workflows without replacing the underlying OCR engine, flagging potentially inaccurate outputs for human review or selecting the best result from multiple OCR systems.

OpenBMB released PilotDeck 2.0 with Tsinghua, ModelBest and AI9Stars on October 18, 2026, an open-source agent operating system with isolated workspaces and editable memory, according to Pandaily.

ModelBest's Funding and Edge-AI Partnerships

ModelBest completed a hundreds-of-millions-yuan financing round in April 2026 jointly led by Shenzhen Capital Group and Huichuan Industrial Investment, with proceeds earmarked for integrating model capabilities into general automation and embodied intelligence product lines, according to akta.pro company records. The same records list a February 2026 round of hundreds of millions of yuan led by China Telecom with strategic collaboration to expand AI R&D across judicial, automotive and education sectors, and a November 2025 strategic partnership with Intel to adapt and accelerate MiniCPM edge models on Intel's latest desktop flagship chips for AI PC applications.

Earlier MiniCPM Releases From ModelBest

ModelBest open-sourced MiniCPM-V 4.7 on October 14, 2026, its first MoE multimodal model, in 35B-A3B and 1B sizes, with the 35B-A3B variant running on an iPhone 17 Pro Max using about 3GB of memory and decoding over 13 tokens per second, Pandaily reported. On August 12, 2026, ModelBest filed A-share IPO counseling documents with the Shanghai Securities Regulatory Bureau, with CITIC Securities as advisor, according to Pandaily. Caixin Global reported the same August 12 filing as pre-IPO tutoring with Beijing's securities regulator for a listing on the Shanghai STAR Market.

The MiniCPM-OCR series is available on GitHub, Hugging Face and ModelScope, and supports vLLM, SGLang, llama.cpp and Ollama, along with SDKs, APIs, Agent Skill plugins and open-source fine-tuning code. Day-0 hardware adaptation covers AMD, Arm, Houmo AI, Intel and NVIDIA, and through the FlagOS software stack the series has Day-0 support for AI and edge chips from CIX Technology, Huawei Ascend, Hygon, Iluvatar CoreX, Kunlunxin, MetaX, Moore Threads, Sunrise, T-Head, TsingMicro, NVIDIA and Apple Mac (Arm).