A preprint finds a unified English hate-speech model led most benchmark comparisons, while cross-lingual performance remained uneven.