<p>Previous research comparing AI-generated and human-written academic abstracts has primarily focused on general linguistic characteristics, overall quality assessment, and rhetorical patterns. No study has systematically compared AI-human texts in terms of three dimensions of lexical complexity (i.e., lexical density, lexical sophistication, and lexical variation), leaving the holistic lexical profile of AI-generated abstracts largely unexplored. To help address the gap, the present study compared lexical complexity differences between ChatGPT-generated and human-written medical abstracts. Two comparable corpora (totaling 600,000 tokens) were compiled, with 800 original abstracts from four prestigious journals and 800 counterparts generated by ChatGPT 4o based on texts of the same set of articles using journal-aligned prompts and exemplar abstracts. Three dimensions of lexical complexity were operationalized by adopting twenty-five measures in Lu’s (<CitationRef CitationID="CR48">2012</CitationRef>) Lexical Complexity Analyzer, and group differences and correlations were evaluated with independent samples t-tests and Pearson correlations. Results showed that ChatGPT-generated abstracts exhibited significantly higher levels of lexical complexity than human-written abstracts on all the three dimensions and all the measures (e.g., more lexical words and sophisticated words, and a broader lexical repertoire). Thus, they are more informationally dense, more specialized, and less repetitious, reflecting more advanced language use and richer vocabulary knowledge. Results also revealed low correlations between human-written and ChatGPT-generated abstracts, indicating their substantial differences in summarizing academic discourse and in making lexical choices. Finally, the study discusses the possible factors influencing these lexical differences in the theoretical context of AI versus human language production mechanisms as well as theoretical and practical implications.</p>