MAKALE ÖZETİ Aim: This study aimed to evaluate and compare the performance of 4 LLMs in responding to common prosthodontic complication scenarios. Methods: An expert panel design was employed to assess the quality of responses generated by the four LLMs across five clinically relevant prosthodontic scenarios. Each scenario was entered 15 times into each model, yielding 300 responses in total. 7 prosthodontists independently evaluated the anonymized responses based on four criteria: clinical accuracy, applicability, clarity, and comprehensiveness, using a 5-point Likert scale. Descriptive statistics, one-way ANOVA, and post-hoc pairwise comparisons were performed to detect inter-model differences. The internal consistency (reliability) of evaluators’ ratings for each model was assessed using Cronbach’s alpha. Results: Significant differences were found among the models for all evaluation criteria (p<.05). ChatGPT achieved the highest mean scores for clinical accuracy (3.89), clarity (3.90), and comprehensiveness (4.09), while Gemini showed the highest applicability (3.88). Cronbach’s alpha coefficients indicated strong reliability across all domains (0.776–0.880), with ChatGPT showing the highest single-domain coefficient (α=0.820–0.880). Conclusion: Among the evaluated models, ChatGPT and Gemini generally received higher ratings on specific criteria in pairwise comparisons; however, the four models should not be interpreted as forming a definitive overall ranking. While LLMs show potential as supportive tools in prosthodontic education and clinical decision-making, their outputs should not replace expert judgment.