The Evolving Capabilities of AI (Image courtesy of MarkerAI)
SEOUL, Dec. 16 (Korea Bizwire) – A domestic AI model has demonstrated remarkable proficiency in South Korea’s notoriously challenging College Scholastic Ability Test (CSAT), achieving a near-perfect score in the Korean language section.
According to the ICT industry on December 16, startup MarkerAI tested OpenAI’s models on the 2025 CSAT, with its “o1-Preview” model earning a raw score of 97, placing it in the top tier.
Tackling South Korea’s Toughest Test
The “o1-Preview” model missed only one question—Question 8, a notoriously difficult item with an 81.5% incorrect response rate. The question involved analyzing two nonfiction passages on modernization and applying logical reasoning to a case study. MarkerAI explained that the AI struggled to grasp the nuanced context and underlying intent of the problem.
Other AI models also attempted the exam, with “o1-Mini” scoring 78 and “gpt-4o” scoring 75, both falling into the fourth tier of rankings. MarkerAI has been testing AI capabilities on CSAT exams over the past decade to evaluate their comprehension and problem-solving skills.
Implications for AI in Education
The success of “o1-Preview” underscores the growing capability of AI to handle complex tasks requiring logical reasoning and contextual understanding. However, it also highlights areas where AI models still struggle, particularly with subtle contextual cues and abstract reasoning—skills critical for human intelligence.
As AI continues to evolve, its ability to perform on high-stakes academic assessments could reshape perspectives on its role in education and testing.
Kevin Lee (kevinlee@koreabizwire.com)
