jina-ocr-v1 is a visual document parser with 3.4B parameters and 570M active, plus speculative decoding. PDFs, scans, tables, charts, and invoices come back as clean markdown.