ChatGPT Told to ‘Solve the CSAT and Give Me the Answers’... Solves Korean, English and Math in 3 Minutes for Perfect Scores
- Input
- 2026-11-20 07:38:25
- Updated
- 2026-11-20 07:38:25

[Financial News] As AI cheating moves to the center of public debate, OpenAI’s ChatGPT and Google’s Gemini solved the Korean, math and English questions (excluding the listening section) from the 2027 College Scholastic Ability Test in about three minutes, scoring full marks in all three.
According to Newsis on the 19th, the PDF of the CSAT exam papers released that day by the Korea Institute for Curriculum and Evaluation was given to GPT-6.2 Astra and Gemini 4 Hydrogen, which were asked to solve the questions. They scored 100 raw points in Korean and math, and earned 63 out of 63 points on the 28 English reading questions, excluding listening.
When the Korean, math and English exam papers were attached separately with the request, “Solve them and list the answers,” both AIs provided answers to all 56 Korean questions and all 46 math questions. They did not use web searches while solving the questions.
Actual test-takers choose just one elective subject in each section and answer 45 Korean questions and 30 math questions. In this test, however, both AIs were made to answer two Korean electives and three math electives. In effect, the AIs answered 11 more Korean questions and 16 more math questions than test-takers do.
GPT-6.2 Astra, with its reasoning level set to “High,” took 44 seconds to provide its Korean answers, 1 minute 25 seconds for math and 35 seconds for English, for a total of 2 minutes 44 seconds. It also used a code execution tool for some calculations.
Gemini also answered every question correctly, but in math it sometimes said it could not accurately read certain graphs or equations and asked for the problem conditions to be checked again. ChatGPT reportedly gave some incorrect answers in an additional test using the same odd-numbered exam form on its paid version.

On the “CSAT LLM Solution Dashboard” run by developer “hehee9,” GPT-6.2 Astra, Sol and Gemini 4 Hydrogen, among others, also answered every common and elective question in this year’s Korean and math sections correctly. Last September, OpenAI Astra scored a perfect 450 overall on the previous year’s CSAT, across Korean, math, English, Korean history and four inquiry subjects, according to the dashboard.
These AI problem-solving abilities are raising concerns amid recent exam-cheating scandals. The top-ranked candidate in this year’s tax accountant exam admitted to cheating using a micro camera and generative AI, among other tools. Allegations that a perfect scorer on the Legal Education Eligibility Test (LEET) used AI to cheat have also prompted a police investigation.
The government has urgently stepped up its response to AI-related cheating concerns that surfaced ahead of the CSAT. At a meeting with relevant ministries on the 16th, Prime Minister Han Seong-suk ordered smart devices to be handed in at test centers and checks to be made for anyone carrying them, and called for a review of how state-administered exams are managed and operated. As a result, emergency disaster text messages aimed at preventing cheating were sent three times during the CSAT on the 19th, for the first time ever.
[email protected] Kim Hee-sun Reporter