
[Diary Point=Reporter Choi Seung-jin] In an experiment using the Korean-language section of South Korea’s 2027 College Scholastic Ability Test (CSAT), the paid version of Gemini scored 100 in about 20 seconds, while Claude Opus 5.5 did so in about two minutes. The paid version of ChatGPT took 6 minutes and 26 seconds to provide answers along with its reasoning. One report said a free version answering the same questions got 13 of 56 wrong.
Results varied by model and testing conditions. The paid Gemini answered all 56 questions correctly: 34 in the common section, plus 11 each from the two elective sections, “Speech and Writing” and “Language and Media.” Claude Opus 5.5, tested by JTBC, also earned a perfect score across all 56 questions, including both electives.
The Korea Economic Daily tested the paid version of ChatGPT, asking it to explain its answer choices. The answers arrived after 6 minutes and 26 seconds, and the outlet confirmed a perfect score against the answer key. In separate tests, the Chosun Ilbo reported that paid Gemini scored 100 in about 20 seconds, while a free version of ChatGPT got 13 questions wrong in 2 minutes and 22 seconds. Since the outlets used different prompts, response requests and models, the times are not directly comparable.
Students have 80 minutes to answer 45 Korean-language questions, choosing one of the two elective sections and completing its 11 questions. The AI experiments used digital files of the publicly released test, and most of the reported results covered both electives. That differs from an exam room, where students receive a paper test, mark an answer sheet and work within the 80-minute limit.
The difficulty of this year’s Korean-language test also drew comment. At a briefing on the exam’s content on the 19th, Han Byung-hoon, an EBS lead instructor for Korean and a teacher at Yesan High School, said, “The overall difficulty of this year’s Korean test was lower than last year’s CSAT and the September mock exam.” He added that “the amount of information in the passages fell considerably, and their structure was less complex.”
AI’s ability to produce answers quickly has also raised concerns about device use during the exam. The Education Ministry said that using a smart device such as smart glasses during the test will invalidate the exam and bar the student from taking it the following year; even bringing in or merely possessing such a device will invalidate that year’s test. The Korea Institute for Curriculum and Evaluation is accepting challenges to the published questions and answers until 6 p.m. on November 23, with final answers due to be released on December 1.
