In the binderyThis page is under construction.
Research
மொழிTamil for “language”
Benchmarks & the state of Arabic data
An open, honest record of how today’s models perform in under-resourced languages, dialect by dialect and domain by domain, and where native data is still missing.
Forthcoming editions
Written by native experts, not translated from English benchmarks.
- Native-written tasksEvery question is written in Arabic by a native-speaking expert.
- Dialect-taggedResults break down by variety, not just “Arabic”.
- Expert-gradedAnswers scored against rubrics written by professionals in the field.
Co-author a benchmark with us
Have a domain where Arabic performance matters? We’ll build an industry benchmark with your team and our experts.