PaddleOCR (92.6K stars) beats MinerU (83.5K) and Tesseract (77.5K) as the most-starred open source OCR project on GitHub. Below are all 138 OCR and document-parsing repos with 1,000+ stars in the Text Mining directory , ranked.
Every RAG pipeline eventually hits the same wall: the PDF. Before a language model can answer anything about your contracts, papers or invoices, something has to turn pixels and layout into clean text. The tools for that job are splitting into two camps. In HackerNoon's comparison of OCR and vision LLMs , template-based OCR breaks when layouts shift, while vision LLMs read documents more flexibly but cost more at volume. The author's verdict: "It could mostly be a hybrid approach."
Meanwhile, the vision models are shrinking. Xiaomi-OCR-0 handles text, tables and formulas with 0.8B parameters, and TeleOCR does structured document parsing with 1.2B.…