Nougat

Nougat, short for Neural Optical Understanding for Academic Documents, is a research model from Meta that parses scientific PDFs into a lightweight markup format. Rather than running conventional OCR, it uses a vision encoder-decoder model built on the Donut architecture to read page images and emit Mathpix-compatible Markdown, preserving LaTeX math, tables, and scientific notation that ordinary text extraction mangles. It works best on English-language papers of the kind found on arXiv and PubMed Central and struggles with non-Latin scripts such as Chinese or Russian. The tool installs from pip as nougat-ocr and runs as a CLI, Python library, or API service, with small and base model checkpoints available; GPU inference is recommended, and CPU runs may need a flag to avoid false failure detection. The code is MIT licensed while the model weights are CC-BY-NC, so the weights are restricted to non-commercial use.