Evaluating generative models is one of the most challenging tasks in the field of AI. Unlike traditional machine learning systems, where performance can often be measured with objective accuracy or error rates, generative models produce outputs that are inherently open-ended. Their success is not defined solely by whether they produce a correct answer, but also by qualities such as fluency, creativity, coherence, fidelity to input prompts, and alignment with human expectations. This multidimensional nature of generative outputs requires a richer set of evaluation methods and benchmarks than those used in conventional supervised learning.