Word Error RateArabic foundation models & open AI
Misraj-DocOCR
A curated, expert-verified Arabic OCR benchmark of 400 pages, with ground truth for text and layout fidelity.
Kawn Lab
Benchmarks
Open benchmarks Kawn Lab built for Arabic tasks that had no reliable way to be measured.
Explore our published benchmarks
CoverageArabic foundation models & open AI
SadeedDiac-25
SadeedDiac-25 unifies Modern Standard and Classical Arabic diacritization in one expert-reviewed benchmark.
Beats GPT-4o miniArabic foundation models & open AI
Tarjama-25
A bidirectional Arabic-English translation benchmark of 5,000 expert-reviewed sentence pairs across five domains.
Get started
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.