MORFES: A Benchmark for Productive Inflectional Competence in Modern Greek

Researchers released MORFES, a 500-item benchmark for testing whether models can correctly recognize and generate Modern Greek inflected forms.

The benchmark emphasizes lower-frequency lemmas to reduce memorization and better measure true morphological competence.

Across several open-weight models, the newly released Sophea-Genesis-1 performed best on the morphology task.

MORFES fills a gap left by fact-focused evaluations and offers a better way to assess grammatical ability in a highly inflected language.