olmOCR 7B is a focused, document-oriented model built to extract and interpret text from images — think scanned PDFs, photographs of pages, or complex layouts. It handles the messy reality of real-world documents where fonts, columns, and image quality vary. As a quantized FP8 variant, it trades a small amount of precision for reduced memory footprint and faster inference.