Adding to the Q: Any good small open-source model with a high correctness of reading/extracting Tables and/of PDFs with more uncommon layouts.

I haven't tried it yet, but I bookmarked this recently: https://github.com/opendataloader-project/opendataloader-pdf