# Unlimited-OCR

> Open-source vision-language model for parsing multi-page documents in one pass

Unlimited-OCR is a 3B parameter vision-language document parsing model developed by Baidu. It processes entire multi-page documents and PDFs in a single pass across a 32k context window, maintaining formatting, reading order, and tables without heavy chunking pipelines. It runs locally and supports deployment via Transformers, vLLM, and SGLang.

Website: https://huggingface.co/baidu/Unlimited-OCR

## What it is for

- Parse multi-page PDF documents into structured text in a single pass
- Extract clean text from dense documents for RAG pipelines without loss of reading order
- Serve high-throughput local document OCR using vLLM or SGLang backends

## Features

- One-shot long-horizon multi-page document parsing
- 32,768 token context length window
- Native support for both single-image high-resolution and multi-image batch modes
- Official inference support with Hugging Face Transformers, vLLM, and SGLang
- Detection marker post-processing for bounding boxes and block separation
- Open weights released under the MIT license

## Pricing

Open source. Free and open source (MIT license)

## Details

- Type: AI model
- Category: [AI](https://xiry.app/topics/ai.md)
- Open source: yes
- Platforms: Linux, API, Code library, Self-hosted
- For: AI engineers, developers, and researchers building document processing and RAG pipelines.
- Alternative to: PaddleOCR, DeepSeek-OCR
- Tags: ocr, vision language, pdf parser, transformers, vllm, rag

## Why it was saved

- [@roddycrypto_, 17 September 2026](https://www.instagram.com/reel/DdX2Nz2tyu9/): Solves context loss and complex chunking pipelines in RAG workflows by reading up to 100+ pages in one pass while preserving reading order, tables, formulas, and formatting locally.
- [@ai.honeycove, 28 July 2026](https://www.instagram.com/reel/DbWZwciSP4l/): Unlike traditional OCR tools that chunk documents page by page and lose context, it maintains document structure, layout, formulas, and tables across long contexts locally without cloud API costs.

## Related

- [NuMarkdown-8B-Thinking](https://xiry.app/resources/numarkdown-8b-thinking): Open-source vision-language model converting complex documents into structured Markdown using reasoning. (AI model, Open source)
- [Reducto](https://xiry.app/resources/reducto): Agentic document parsing and extraction platform for AI pipelines. (Web app, Freemium)
- [stepfun-ai/Step3-VL-10B](https://xiry.app/resources/stepfun-ai-step3-vl-10b): Open 10B multimodal model with parallel reasoning competitive with frontier systems (AI model, Open source)
- [WeVisDoc](https://xiry.app/resources/wevisdoc): Open-source vision-language model converting complex document images to structured Markdown (AI model, Open source)
- [ai-engineering-hub](https://xiry.app/resources/ai-engineering-hub): Open-source tutorials and project templates for LLMs, RAG, and AI agents. (GitHub repo, Open source)
- [Gemini Notebook](https://xiry.app/resources/gemini-notebook): AI research assistant grounded strictly in your uploaded notes and documents. (Web app, Freemium)

Updated 3 October 2026. Page: https://xiry.app/resources/unlimited-ocr

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.