PaddleOCR

PaddleOCR is the workhorse of open source OCR. It comes out of Baidu's PaddlePaddle ecosystem and reads text from images and documents in over 100 languages. The newer pieces go further: PP-StructureV3 pulls layout and tables into Markdown or JSON, and PaddleOCR-VL is a small vision-language model that parses whole documents including formulas, tables, seals, and charts. Apache 2.0, free, and cleared for commercial use.

It runs on CPU or GPU, so you do not strictly need a graphics card to start, though a GPU helps at volume. The PP-OCRv6 models claim a fivefold CPU speedup over v5 with better accuracy, and PaddleOCR.js runs the older PP-OCRv5 model right in the browser. No server at all is also an option: the official site offers a free hosted API, capped at 3,000 pages per model per day, with a Baidu AI Studio token.

For most teams replacing a paid OCR API, this is the first tool to try. It matches the core job: text, tables, layout. Solo to large teams: free. docling-project/docling is the alternative when the goal is feeding documents to an LLM.

The catch is complexity. It is a big toolkit with many models, versions, and configuration options, and the docs lean Chinese-first in places. The repo has also been quiet since its June 2026 release. Budget the setup time.