Unlimited OCR is an open-source document-parsing model from Baidu designed for one-shot, long-horizon OCR. Instead of chopping long documents into fragile page-by-page steps, it parses lengthy, complex documents in a single pass, handling dense layouts, tables, and multi-column text. Released on GitHub with an MIT license, it has gained over 28,000 stars and was updated in October 2026.
Traditional OCR pipelines struggle with long documents: errors compound across pages, reading order breaks, and tables fall apart. Unlimited OCR tackles this with an architecture built for long-horizon parsing, so contracts, reports, books, and scanned archives can be converted into structured text in one shot. The model weights are published openly, and the model is also available through Hugging Face and Baidu Cloud for teams that prefer a hosted route. It integrates with popular inference tooling such as vLLM, making local deployment practical.
Unlimited OCR is free and open source. Downloading and running the model yourself costs nothing beyond your own hardware; optional hosted access through Baidu Cloud follows that platform’s own pricing.
Unlimited OCR is for developers and data teams who process large volumes of long documents: researchers digitizing archives, engineers building document AI pipelines, and anyone frustrated by page-by-page OCR errors on contracts, manuals, and reports.