DeepSeek-OCR
DeepSeek-OCR is an open source model for Contexts Optical Compression, built to explore the boundaries of visual-text compression and investigate the role of vision encoders from an LLM-centric viewpoint. It is designed to compress long contexts through optical 2D mapping, using DeepEncoder as the core engine and DeepSeek3B-MoE-A570M as the decoder. DeepEncoder maintains low activations under high-resolution input while achieving high compression ratios, keeping the number of vision tokens manageable for document understanding. The model supports OCR and document parsing workflows for images and PDFs, with inference through vLLM or Transformers. Users can run image OCR with streaming output, process PDFs with high concurrency, or run batch evaluation for benchmarks. DeepSeek-OCR can convert documents to Markdown, perform free OCR without layouts, parse figures, describe images in detail, and locate referenced text inside an image.
Learn more
Mistral Document AI
Mistral Document AI is an enterprise-grade document processing solution that combines advanced Optical Character Recognition (OCR) with structured data extraction capabilities. It achieves over 99% accuracy in extracting and understanding complex text, handwriting, tables, and images from various documents across global languages. It can process up to 2,000 pages per minute on a single GPU, offering minimal latency and cost-efficient throughput. Mistral Document AI integrates OCR with powerful AI tooling to enable flexible, full document lifecycle workflows, making archives instantly accessible. It supports annotations, allowing users to extract information in a structured JSON format, and combines OCR with large language model capabilities to enable natural language interaction with document content. This allows for tasks such as question answering about specific document content, information extraction, and summarization, and context-aware responses.
Learn more
Docling
Docling is an easy-to-use, self-contained, MIT-licensed open source toolkit for converting messy documents into structured data and simplifying downstream document and AI processing. It can parse many popular document formats into a unified and richly structured Docling Document, including PDF, DOCX, PPTX, XLSX, HTML, Markdown, AsciiDoc, CSV, images, audio, and scanned pages through an OCR engine of the user’s choice. Docling detects tables, formulas, reading order, chunks, bounding boxes, page headers and footers, pictures, captions, code, list items, paragraphs, cells, and document structure, making extracted content easier to process, search, and ingest into AI, RAG, and agentic systems. It can export parsed documents to JSON, text, Markdown, HTML, and Doctags, giving developers flexible outputs for pipelines and applications. Docling stores and traverses components according to reading order, partitions documents into bite-sized contiguous text chunks.
Learn more
Mistral OCR 3
Mistral OCR 3 is the third-generation optical character recognition model from Mistral AI designed to achieve a new frontier in accuracy and efficiency for document processing by extracting text, embedded images, and structure from a wide range of documents with exceptional fidelity. It delivers breakthrough performance with a 74% overall win rate over the previous generation on forms, scanned documents, complex tables, and handwriting, outperforming both enterprise document processing solutions and AI-native OCR tools. OCR 3 supports output in clean text, Markdown, or structured JSON with HTML table reconstruction to preserve layout, enabling downstream systems and workflows to understand both content and structure. It powers the Document AI Playground in Mistral AI Studio for drag-and-drop parsing of PDFs and images and integrates via API for developers to automate document extraction workflows.
Learn more