Examines how model size affects essay scoring accuracy in teenage English learners, suggesting larger models yield better alignment with human judgments.
The utility of generative Artificial Intelligence (AI) in language assessment has become an important topic in recent research. Due to its sophisticated linguistic processing capabilities, it has been expected to provide support in automated essay scoring. Although existing studies confirm the potential of this technology, they had not previously examined whether, or how, model size influences generative AI performance. Our study addresses this gap by analyzing the Intraclass Correlation Coefficients and Pearson correlations between scores assigned by humans and those assigned by three models within the GPT-5 family (GPT-5 , GPT-5 mini, GPT-5 nano) across 4,493 essays written by teenage learners of English with various language backgrounds. The results clearly show that as the generative AI model scales up from GPT-5 nano to GPT-5, its alignment with human judgment becomes stronger. This finding indicates the model size of GPT-5 as a critical factor in determining its performance, leading to the recommendation that researchers and practitioners utilize the largest available models for reliable automated essay scoring. Keywords—generative AI, GPT-5, model size, automated essay scoring, assessment
No takes yet. Share an insight, caveat, or question.
Haerim Hwang (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: