Specialized NLP
1.5B
Open weights
Sadeed
Arabic diacritization built on Kuwain, restoring full Tashkeel for MSA and classical Arabic.
Arabic diacritization model built on Kuwain 1.5B. Adds Tashkeel (vowel marks) back into raw Arabic text. Context-aware diacritization that delivers accuracy matching proprietary LLMs on equivalent domains. Useful as preprocessing for TTS, search indexes, and NLP pipelines.
Benchmarks
| Dataset | Metric | Result |
|---|---|---|
SadeedDiac-25Diacritization results were being reported on narrow, inconsistent test sets, which made models hard to compare honestly. A single benchmark covering both registers and several genres is what lets one diacritization result be read against another. | Coverage | MSA + Classical Arabic |
Access
Cloud API
On-premises
Let's talk about what you're trying to build.
Tell us the problem. We'll tell you honestly whether AI is the right answer.