Unlimited-OCR-2 (Baidu)

What is Unlimited-OCR-2 (Baidu)

Unlimited-OCR-2 is a 3B-parameter OCR model open-sourced by Baidu for document recognition, and it can be deployed locally. The weights are freely available under the MIT license on Hugging Face, where the model was created on 2026-10-26.

How to use

1. Install dependencies, e.g. pip install torch torchvision pillow transformers==4.52.1 huggingface-hub tokenizers addict einops easydict (tested with Python 3.10 on an NVIDIA A800-80GB).
2. Load the model with AutoModel.from_pretrained("baidu/Unlimited-OCR-2", trust_remote_code=True) and its tokenizer.
3. For a multi-page document, pass all page images in order to model.infer_multi(...) so the model can use cross-page context.
4. For a single page, call model.infer(tokenizer, image_file=...).
5. For serving, use SGLang (recommended) or vLLM.
6. Or try the online demo on Hugging Face Spaces (baidu/Unlimited-OCR-2), ModelScope or Baidu Cloud.

Use cases

• Parsing ultra-long documents of hundreds of pages in one pass Multi-page OCR that uses cross-page context Converting scanned documents and page images to text Self-hosted OCR serving with SGLang or vLLM Benchmarking long-document parsing (e.g. MarathonDocBench)

Price from official sources

Free

• Open-weight model released under the MIT license;
• free to download from Hugging Face.

Check the official site for current pricing · Price not yet checked by our editors

Source: huggingface.co · Info as of Nov 2, 2026

Company

Name

Baidu Inc.

Social

• x
• github