<p>In this work, we propose an entropy-based framework to quantify uncertainty in Generative Pre-trained Transformer (GPT) models when extracting mathematical equations from images of varying resolutions and converting them into mathematical notation. By measuring the conditional entropy of the model’s output token sequences, we assess how confidently GPT recognizes and converts mathematical expressions into LaTeX format. Our experimental results, obtained using a custom Python implementation that we developed and made publicly available on GitHub, reveal an inverse relationship between image clarity and model uncertainty: high-resolution images yield lower entropy values and more accurate LaTeX outputs, while low-resolution inputs increase entropy and error rates. These findings demonstrate how entropy analysis, grounded in information-theoretic concepts, can effectively quantify GPT model uncertainty in real-world tasks.</p>