The growing adoption of Large Language Models (LLMs) necessitates rigorous ethical evaluation, especially in linguistically diverse contexts like the Spanish-speaking community. This study systematically evaluates gender biases in the justifications generated by GPT-3.5-turbo, GPT-4o-mini, Claude-Haiku, and Llama3-Groq. Through a structured set of questions, subcategories of sexist biases related to emotions, leadership, occupations, and social roles were analyzed. The methodology integrated Natural Language Processing (NLP) techniques: semantic similarity analysis (SBERT), lexical co-occurrences, polarity assessment, fine-grained emotion classification, and topic modeling (LDA). The results reveal consistent patterns of gender stereotype reproduction across all models, albeit with variations in their manifestation. This research provides empirical evidence of the prevalence of gender biases in contemporary LLMs when generating content in Spanish and offers crucial insights for future mitigation strategies.