Benchmark
In Arabic: الاختبار المعياري
A standard test used to compare AI models, for example on maths problems or Arabic reading. Useful but limited: a model can score well on a test and still struggle with real tasks. Companies' benchmark claims should be read as their claims.