Large language models can now produce remarkably fluent literary translations, but fluency does not necessarily entail the preservation of a text’s distinctive voice, interpersonal meanings, or pragmatic effects. This study examines this tension through a qualitative case study of an AI-generated Persian translation of O. Henry’s The Gift of the Magi. Drawing on House’s (2015) Revised Translation Quality Assessment Model, the study constructs independent textual profiles of the English source text and Persian target text and compares them across field, tenor, mode, genre, and textual function. The analysis reveals a high degree of functional correspondence; the translation preserves the story’s propositional content, narrative progression, thematic development, interpersonal relationships, cultural references, and overall communicative function. The principal divergences emerge not from mistranslation or loss of meaning but from stylistic and pragmatic shifts. In particular, explicitation, normalization, formalization, syntactic smoothing, and moderate emotional intensification occasionally make the Persian text more explicit, literary, and emotionally expressive than the original, reducing aspects of O. Henry’s conversational informality, understated irony, and stylistic restraint. The results highlight a central challenge in AI-generated literary translation: semantic and functional adequacy can coexist with subtle changes in literary voice. The findings are particularly relevant to translation scholars, literary translators, and educators seeking to evaluate and critically use AI-generated literary translations.