LlamaIndex

LlamaIndex delivers the world’s most accurate agentic OCR and document-specific AI workflows, powering complete enterprise automation.

On HuggingFace and in our open-source repos, we are pushing forward document parsing, OCR, and extraction with several projects:

• ParseBench -- A comprehensive dataset for evaluating document parsing pipelines and OCR models
• ExtractBench -- A benchmark dataset for evaluating schema-guided extraction from enterprise documents
• ExtractBench Pro -- A challenging benchmark dataset for evaluating long-tail document extraction
• LiteParse -- Our lightweight, open-source document parser. Handles multiple formats, runs locally, and integrates with any OCR model
• Visual Document Retrieval -- An exploration into training models purely for document screenshot retrieval

LlamaParse

LlamaParse provides end-to-end document understanding with AI-powered parsing, extraction, and indexing. Transform complex layouts, tables, and handwriting into actionable insights with industry-leading accuracy.

• Parse -- Parse any document, any format, with SOTA accuracy
• Extraxt -- Extract structured data. using your own schemas, from any document
• Classify -- Classify documents into a subset of classes
• Split -- Split documents according to specific categories
• Index -- Index and retrieve your data