Unlimited-OCR — 3B OCR That Parses 100-Page PDFs in One Shot
Post by Ahmed Islam — China open-sourced a peanut-sized OCR
The Problem
Every other OCR tool chops your doc into pages and loses the thread. This reads the whole thing in a single pass.
Key Features
- One-shot “long-horizon” parsing — 32K context window
- Multilingual, out of the box
- 93% on standard parsing benchmark (+6 over baseline)
- <0.11 error rate past 40 pages
- Runs 100% locally on your own hardware
Compatible Platforms
Transformers, vLLM, SGLang, Docker, Ollama, llama.cpp
Cost Comparison
| Service | Cost per 1,000 pages |
|---|---|
| AWS Textract / Google Vision / Azure AI Doc Intelligence | 1.50 – 15 |
| Unlimited-OCR (local) | Free. Forever. |
Background
Built by Baidu to push DeepSeek-OCR further. Already 1.9M downloads on Hugging Face.
100% open source. Only 3B parameters.