Research on the effectiveness of GPT-4o and GPT5.4 models in solving olympiad programming tasks on the Eolymp platform

The object of the research is the process of solving Olympiad programming tasks with Ukrainian statements from the Eolymp platform using the large language models GPT-4o and GPT-5.4. As a result we discovered that LLM average efficiency in solving sports programming tasks with Ukrainian statements was nearly identical for C++ and Python, averaging around 32% for GPT-4o and 56% for GPT-5.4. It was revealed that LLMs effectiveness when providing Ukrainian and English statements for the same tasks was approximately the same (p>0.05). It was found that the proportion of tasks completely solved by LLMs was about 13% for GPT-4o and 37 % for GPT-5.4 of the total number considered in this study. The result for GPT-4o is comparable with the results of solving similar tasks with English statements on the Codeforces platform, while the result for GPT-5.4 is significantly larger, however, the number of tokens used by the new model GPT-5.4 is almost 3 times larger.