: Includes a collection of high-quality Arabic fonts, such as Naskh, Thuluth, Kufi, and Diwani.
The model has been fine-tuned on Al-Mutanabbi and Nizar Qabbani . It can mimic classical Qasida structures (metered poetry) or generate modern romantic prose. Unlike standard models that output generic rhymes, V3.1 R1 37 respects Al-Bahr (poetic meter) like Al-Tawil or Al-Kamil . Al Rassam Al Arabi V3.1 R1 37
Many historical Arabic documents are degraded scans. While not a vision model natively, V3.1 R1 37 pairs with OCR tools to extrapolate missing text. If an OCR reads "ذهب الـ... إلى السوق" (Went the _ to the market), the model predicts "ولد" (boy) or "رجل" (man) with 98% contextual accuracy. : Includes a collection of high-quality Arabic fonts,
The model scores 84.6% . Comparatively, GPT-3.5 scores 49%, and GPT-4-Turbo scores 71%. This suggests that V3.1 R1 37 is exceptionally good at mathematical reasoning, logic puzzles, and legal interpretation when the prompt is in Arabic. Unlike standard models that output generic rhymes, V3
