Best OCR and vision language models you can run locally that transform documents, tables, and diagrams into flawless markdown copies with benchmark-crushing accuracy.