MiniCPM-OCR 1.0

Use it to recognize and parse documents, convert their contents into structured formats, or extract specified fields such as receipt totals.

Availability

BF16 checkpoint available on Hugging Face

Pricing model

Not established

Deployment

Usage with vLLM

Privacy

Not established

Licence

Apache License 2.0

• The model handles text, tables, formulas, charts, seals, and key information extraction. ^{[1] [2]}

  | Tasks | Text, tables, formulas, charts, seals, and key information extraction |

• It supports Chinese and English documents, including PDFs, scans, photographs, skewed pages, and recaptured documents. ^{[1] [2]}

  Handles Chinese and English documents and PDFs, including scans, curved pages,

• The model weights are available on Hugging Face and ModelScope. ^{[1] [2]}

  Model weights, code, and documentation are released under the

• Deployment guides are provided for vLLM, SGLang, llama.cpp, Ollama, Transformers, and FlagOS. ^[1]

  - 🛠️ **Easy to deploy and develop**: **vLLM, SGLang, llama.cpp, and Ollama from day one**, with

• The model has 1.3 billion parameters. ^{[1] [2]}
• Model weights, code, and documentation are released under the Apache License 2.0. ^[1]

  Model weights, code, and documentation are released under the

• The draft head speeds up decoding by proposing tokens that the main model verifies in one pass. ^[3]

  **DSpark** speeds up decoding with a small draft model that proposes tokens the main model then

• With this head, MiniCPM-OCR 1.0 improves end-to-end processing efficiency by 55%. ^{[2] [3]}

  With this head, MiniCPM-OCR 1.0 improves **end-to-end processing efficiency

• The repository provides a BF16 draft checkpoint for MiniCPM-OCR 1.0. ^[3]

  > **This repository hosts the DSpark draft (speculative decoding) head for

• The usage instructions show serving MiniCPM-OCR 1.0 with DSpark in vLLM. ^[3]
• DSpark requires a matching main and draft model from the same release. ^[3]

  DSpark requires a **matching main and draft model from the same release**; do not pair this draft

• Model weights, code, and documentation are released under Apache License 2.0. ^[3]

  Model weights, code, and documentation are released under the

• New models bring live speech translation and voice features Daily brief 2026-10-28