document understanding / OCR-free document parsing
Donut is an OCR-free end-to-end Transformer model for visual document understanding, including classification and key information extraction, with a synthetic document generator for pre-training.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
choose when you need a layout-aware model that works with OCR token boxes and can adapt to low-resource languages without relying on OCR-free end-to-end training.
Official PyTorch implementation of LiLT: A Simple yet Effective Language-Independent Layout Transformer for Structured Document Understanding (ACL 2022)
choose when you need retrieval-augmented document QA over multiple pages and can accept a retrieval pipeline instead of a single OCR-free transformer.
[CVPR2025] VDocRAG: Retirval-Augmented Generation over Visually-Rich Documents
choose when you need a full document-AI pipeline with layout and OCR components rather than a single OCR-free model.
A Repo For Document AI
choose when you want an ETL pipeline to turn documents into structured data for LLMs, using a broader toolset rather than an OCR-free model.
Convert documents to structured data effortlessly. Unstructured is open-source ETL solution for transforming complex documents into clean, structured formats for language models. Visit our website to learn more about our enterprise grade Platform product for production grade workflows, partitioning, enrichments, chunking and embedding.
choose when you need a fast, lightweight document parser for structured output and can trade OCR-free end-to-end learning for a more modular pipeline.
A fast, helpful, and open-source document parser
choose when you need a vision-language model that handles broader tasks and can be fine-tuned on document data, at the cost of a much larger model.
[ECCV 2024] Official code implementation of Vary: Scaling Up the Vision Vocabulary of Large Vision Language Models.
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
choose when your task is narrowly key-value extraction from form-like documents and you need a specialized, non-OCR-free method.
[MM'2024] PEneo, an effective algorithm for key-value pair extraction from form-like documents, designed for real-world applications.
choose when you need an older OCR-based key information extraction model and can accept its narrower, less general design.
An unofficial PyTorch implementation of "Lin et al. ViBERTgrid: A Jointly Trained Multi-Modal 2D Document Representation for Key Information Extraction from Documents. ICDAR, 2021"
Does the same job, but ships as an app. Useful to a person, not swappable into a codebase.
choose when you want a turnkey OCR/extraction tool based on vision models, rather than a model you train or embed.
OCR & Document Extraction using vision models
choose when you want a simple utility to convert PDFs to Markdown via vision-LLM APIs, without integrating a model yourself.
PDF to markdown using vision LLMs — tables, layouts, and structure preserved
These projects were analysed and named donut among their alternatives. The relationship is not symmetric — how donut rates them is a separate judgement, made when donut is analysed in its own right.
calls donut “OCR-free document understanding transformer, same LMM-based approach”
A lightweight LMM-based Document Parsing Model
calls donut “OCR-free encoder-decoder; choose if you want a smaller model that skips explicit OCR tokens.”
Unlimited OCR Works: Welcome the Era of One-shot Long-horizon Parsing.
calls donut “OCR-free document understanding transformer, choose for structured document parsing rather than raw text recognition”
Official code implementation of General OCR Theory: Towards OCR-2.0 via a Unified End-to-end Model
calls donut “Use for OCR-free document understanding with a Transformer, avoiding traditional OCR pipelines.”
GLM-OCR: Accurate × Fast × Comprehensive
calls donut “if you need an OCR-free document understanding transformer for research or specific image inputs”
Get your documents ready for gen AI
calls donut “An older OCR-free document understanding transformer; choose for general-purpose document parsing research, but it's unmaintained and lacks Nougat's academic focus.”
Implementation of Nougat Neural Optical Understanding for Academic Documents
calls donut “Use only if you want an older, unmaintained OCR-free document understanding model for research; previous generation.”
The end of web parsing. The beginning of scalable pixel-native search. link: https://pixelrag.ai/
calls donut “choose when you need an older lighter OCR-free VLM plus a synthetic document generator, with MIT license.”
Multilingual Document Layout Parsing in a Single Vision-Language Model
calls donut “choose it if you need a historic OCR-free document understanding model for research; it is not a production-ready replacement.”
Ingest, parse, and optimize any data format ➡️ from documents to multimedia ➡️ for enhanced compatibility with GenAI frameworks
calls donut “research transformer for OCR-free document parsing, not production-ready and unmaintained.”
Transforms PDF, Documents and Images into Enriched Structured Data