Description
MiniCPM-OCR 1.0 is the first OCR-specific model in the MiniCPM-V/o family, focusing on text-centric document understanding. It is designed to handle a variety of document types, including text, tables, formulas, charts, seals, and structured information extraction. The model is particularly effective in parsing real-world documents, achieving a state-of-the-art score of 93.68 on the Real5-OmniDocBench benchmark. This makes it a leading choice for document parsing tasks, especially in handling both Chinese and English documents and PDFs, including scans, curved pages, photographs, skewed pages, and recaptured documents.
Read More
The model offers efficient inference capabilities, with DSpark decoding acceleration improving end-to-end processing efficiency by 55%. It boasts a throughput of about 1.49 times that of PaddleOCR-VL-1.6 and 3.62 times that of MinerU2.5 Pro. MiniCPM-OCR 1.0 is easy to deploy and develop, supporting vLLM, SGLang, llama.cpp, and Ollama from the outset. It also offers quantized variants and multi-chip builds, allowing for further fine-tuning. The model ships with a pip SDK, a full API, and agent workflows, making it accessible for developers.
The model architecture includes a SigLIP vision encoder and a Qwen3.5 language model, trained for OCR on top of MiniCPM-V 4.6. It supports tasks such as text, tables, formulas, charts, seals, and key information extraction. The parsing pipeline integrates MiniCPM-OCR 1.0 with the PP-DocLayoutV3 layout model and the minicpmvocr SDK. This setup allows for efficient document parsing, where the SDK parses whole documents and recognizes regions concurrently, assembling results into Markdown in reading order.
MiniCPM-OCR 1.0 is released under the Apache License 2.0, making it open source and accessible for further development and integration. It is particularly suited for industries requiring advanced document parsing and information extraction capabilities, such as finance, legal, and education.
- CategoryAI Models & LLMs
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsWeb
MiniCPM-OCR 1.0 Highlights
OCR-specific model
Text-centric document understanding
State-of-the-art document parsing
Efficient inference with DSpark
Supports Chinese and English documents
Easy deployment and development
Quantized variants and multi-chip builds
Open source under Apache License 2.0
Getting Started with MiniCPM-OCR 1.0
Access model: Visit Hugging Face
Authenticate: Set up API access
Set up environment: Install SDK
Integrate via API: Use provided endpoints
Optimize: Fine-tune for specific tasks
MiniCPM-OCR 1.0's Use Cases
- Document Parsing
- Information Extraction
- Multilingual Support
- Real-time Processing
- Open Source Development










