Abstract
Evaluating student writing is a crucial yet time-consuming process. Large Language Models (LLMs)and other recent advancements in generative AI provide potential solutions to support teachers in essayassessment. This study psychometrically evaluates large language models such as GPT-5.2 and Gemini withhuman raters to examine reliability of these solutions. All 55 essays were collected and assessed by GPT 5.2,Gemini and an experienced teacher using a four-criterion scoring rubric. The aim of this study is to psychometrically compare the results of human raters, GPT-5.2, and Gemini assessments on English essays,to determine their rating reliability, intercorrelation, rating differences, and the construct validity of the ratingrubric. The results demonstrated that GPT-5.2 assigned the highest vocabulary scores, while Geminiproduced the highest grammar scores. Overall, GPT-5.2 showed high agreement with teacher ratings and canbe used as a reliable adjunct to essay grading. However, teacher professional judgment remains importantwhen assessing complex content and contextual features. These findings support the use of generative AI as asupplementary tool for improving the efficiency and consistency of essay assessment in EFL contexts.