ColPali is an openly released model that searches documents by looking at images of their pages directly, instead of first extracting the text.
In plain words
Normal document search pulls text out of each page, often through slow and fragile steps, and loses layout, figures and tables. ColPali, a vision-language model built on PaliGemma, turns each page image into many small embeddings. At search time it matches a text query against them ColBERT-style, token by token. The authors found this simpler and faster than text-extraction pipelines and usually more accurate.
A quick example
Imagine searching a pile of scanned slide decks for a revenue chart. ColPali can match the query to the chart's page from its visual content, even if text extraction would have missed the numbers inside the image.
Where you'll meet it
- RAG over visually rich documents like slides, reports and forms.
- Search over scanned pages.
Related
Sources
- PaperColPali: Efficient Document Retrieval with Vision Language Models, Faysse et al. (arXiv) · read 29 Sept 2026
- Docsvidore/colpali model card, ViDoRe (Hugging Face) · read 29 Sept 2026
- DocsColPali repository README, Illuin Technology (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.