Evaluating the performance of 3 large language models in higher education essay-like assessments in 2024 and 2026

Recent advances in artificial intelligence, especially generative large language models (LLMs), have transformed the higher education sector, raising concerns with academic integrity. The current literature lacks direct comparative analyses between LLMs. In the current study, we evaluated and compared the performance of ChatGPT, Gemini and Copilot (free versions) in 2024 and 2026, following prompts related to coursework essay assessments in computer science education or biomedical science. Our results indicate that LLMs struggle to abide by the word count beyond 1000 words, with Gemini presenting greater deviations. Copilot presented the lowest frequency of reference hallucinations. Overall performance of the 3 LLMs did not reveal any statistically significant differences. Quality assessments of the outputs revealed issues with content and criticality for all the LLMs. Similar performances were observed in 2024 and 2026.