Baidu’s Unlimited-OCR: Open-Source 3B Model That Parses 40+ Page PDFs in One Pass

🚨 CHINA JUST OPEN SOURCED AN OCR THAT EATS 100 PAGE PDFS IN ONE PASS
It’s called Unlimited OCR. Only 3B params. Runs locally.
Every other OCR tool chops your doc into pages and loses the thread. This one reads the whole thing in one continuous pass.
→ One shot full context parsing (32K context window)
→ Multilingual out of the box
→ 93% on the standard parsing benchmark (6 points over baseline)
→ Under 0.11 error rate past 40 pages
→ Runs 100% locally on your own hardware
→ Works with Transformers, vLLM, SGLang, Docker, Ollama, llama.cpp
Traditional cloud OCR (Textract, Google Vision, Azure Doc Intelligence) costs $1.50 to $15 per 1,000 pages.
This runs on your machine. For free. Forever.
Baidu built it explicitly to push DeepSeek OCR one step further. Already at 1.9M downloads on Hugging Face and most people have no idea it exists yet.
100% open source.

Baidu’s Unlimited-OCR is an MIT-licensed 3B-parameter mixture-of-experts model (about 500M active) built on DeepSeek-OCR. It parses multi-page documents in a single forward pass with a 32K context window, using a sliding-window attention design that keeps memory flat as output grows. It scores 93.23 on OmniDocBench v1.5, and the paper’s long-document test uses an in-house set of books, documents and papers from 2 up to 40+ pages. Its text edit distance stays under about 0.11 at 40+ pages, though it rises with length. It runs locally through Transformers, vLLM, SGLang, Docker, Ollama and llama.cpp. Treat 40+ pages as the tested range, and benchmark it on your own documents (especially scans or handwriting) before relying on it