The advent of advanced Large Language Models (LLMs) like GPT-4o and Gemini

1.5 Pro demonstrates significant potential for enhancing problem-solving in

coding. This master’s thesis provides a quantitative comparison of these two

models, focusing on their performance in solving algorithmic and data structure

problems while exploring their reasoning capabilities through experimental

research.

The study adopts a quantitative approach, presenting the models with a

diverse set of coding problems sourced from LeetCode, an online platform for

coding challenges. Each solution is rigorously evaluated through unit testing

to determine correctness and by measuring execution time to assess efficiency.

This comprehensive method offers insights into the models’ problem-solving

effectiveness and computational efficiency.

The findings reveal that GPT-4o outperformed Gemini 1.5 Pro in problemsolving

ability, particularly in handling complex algorithmic tasks. However,

Gemini 1.5 Pro demonstrated slightly better efficiency, achieving marginally

lower execution times on average. Both models significantly outperformed human

programmers in efficiency, producing more optimized solutions. These

results highlight the potential of LLMs like GPT-4o and Gemini 1.5 Pro to

address specific coding challenges effectively, paving the way for their broader

application in computational problem-solving.