AI · AI model
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.
What it's 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
- Topic
- Document Parsing
- 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_: 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: 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
Open-source vision-language model converting complex documents into structured Markdown using reasoning.
Agentic document parsing and extraction platform for AI pipelines.
Open 10B multimodal model with parallel reasoning competitive with frontier systems
Open-source vision-language model converting complex document images to structured Markdown
Open-source tutorials and project templates for LLMs, RAG, and AI agents.
AI research assistant grounded strictly in your uploaded notes and documents.
Updated