olmocr

Why we included this project

Feeding PDFs into an LLM pipeline usually means fighting the layout: text buried in scans, multi-column pages, or headers that repeat on every sheet. olmOCR runs a 7B vision-language model on your own GPU and turns PDFs, PNGs, and JPEGs into clean Markdown, keeping a sensible reading order through figures, tables, equations, and handwriting while dropping headers and footers. Because there is no per-call API fee, the cost stays around $200 per million pages, which makes bulk jobs like building a training corpus or digitizing an archive feasible. It also includes a benchmark suite with thousands of documents, so you can check its output against other OCR tools before you commit. For teams that need large volumes of clean document text without paying per page, this is a solid self-hosted option.