The study evaluates the performance of four large lingual models (GPT-4o, GPT-o3, GPT-4.1 and Claude 3.7 Sonnet) as aids in the selection of parameters of the first trial lens in orthokeratology for the control of myopia. A retrospective analysis used these models to assess refractive error cases. Subjective evaluation included accuracy and quality of answers, objectively focused on differences in lens parameters. GQS and accuracy differed significantly between models [χ²(3) = 39.85, p < 0.001; Kendall's W = 0.148]. GPT-o3 and GPT-4o performed best with GQS of 4.66 ± 0.48 and 4.47 ± 0.5, respectively, scoring well in 83.3% and 76.7% of cases. Errors in the parameters decreased after two correction rounds, mainly focused on BC and RZD. Bland-Altman analyzes showed that most observations fell within the 95% limits of agreement. LLMs can support decision making in CRT, but parameter selection requires physician validation due to biases in BC and RZD.