# Whats the best open source OCR for batch processing pdfs in 2026? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/86bf084e6a7bd05c

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-11-02 05:48 ingested 2026-11-03 02:42

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xlp8nd/whats_the_best_open_source_ocr_for_batch/)

## retained text



```
trying to settle on a fully local setup for batch processing pdfs. like each file is around 20-25 pages and mostly text, a few tables and now and then a page thats just a photo of paper

the shortlist so far

\- docling (and the granite-docling model)  
\- glm-ocr  
\- paddleOcr-vl  
\- olmocr 2  
\- mineru

we considered llamaparse for this kind of stuff but the company now wants every file to stay on our own servers and their self hosted route is an enterprise conversation. liteparse from the same team runs local and is quick on the plain text pages but the markdown is rule based with tesseract as the fallback so merged cells and multi column pages dont always come back clean

is a VLM overkill for mostly text pdfs or does it make more sense to send only the table and image pages to one and keep the rest on a text layer parser? textract and docuent AI are out anyway for the same reason

for the vlm ones, do people feed a whole 25 page pdf in one go or split it into page batches first? splitting every file is the part id rather avoid

Edit: Adding from the comments: unlimited-ocr-2, minicpm-ocr-1.0, surya-ocr-2.5, indra-ocr and xiaomi-ocr-0. mostly curious how these would perform on mixed 20-25 page files
```



document fetched 2026-11-03 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1xlp8nd/whats_the_best_open_source_ocr_for_batch
