Qari-OCR
Small open Arabic vision-language OCR that keeps full diacritics on print
Visit website →About
Qari-OCR is an Arabic-specific OCR model from the NAMAA group at Prince Sultan University in Riyadh, built by fine-tuning a small open vision-language model on synthetic Arabic pages with varied fonts, sizes and full diacritics. It converts a page image directly to text, keeping tashkeel, and runs on a single consumer GPU or a free notebook. The weights are published on Hugging Face under the terms of the base model.
The June 2025 paper reports a character error rate of 0.061 and a word error rate of 0.160 on two hundred fully vocalised printed pages, against 0.436 for Tesseract and 0.210 for Mistral OCR on the same set, and the authors state that handwriting capability is limited because training focused on digital fonts. An independent 2026 study found it degrades sharply on real handwriting and historical manuscripts. The use case is precise: Arabic religious, literary and educational print where diacritics must survive, processed on your own hardware. Free.
AI-assisted draft, human-reviewed before publishing — see how we choose & review tools.
Why we picked it
It is the only OCR we found, open or commercial, that was built around Arabic diacritics and published its error rates against named baselines, and it comes from a Riyadh university. We list it for the vocalised-text niche where the global tools quietly drop the marks.