AI aces Korea's college entrance exam in under three minutes as cheating fears mount
Several AI models scored perfect or near-perfect marks on the Suneung, South Korea's college entrance exam, which was sat just this week. ChatGPT finished the Korean, math and English reading sections in a combined 2 minutes 44 seconds. On exam day, officials also caught 98 cheating cases, some involving smart glasses and hidden cameras. The question of whether exams can still be fair in the age of AI has become a national issue.

What's going on
South Korea held the Suneung for the class of 2027 on November 19, 2026. It's the country's national college entrance exam, and it decides who gets into university and, often, how a life turns out. The state exam authority, the Korea Institute for Curriculum and Evaluation, published the questions the same day. Soon after, journalists and a student fed the exam to major large language models. The best of them got perfect scores, in minutes rather than hours.
The results landed in the middle of a wave of cheating cases, with test-takers caught carrying high-tech gadgets, and the tragic death of a young man accused of AI-assisted cheating on a different exam. Korean outlet Topstar News sums it up this way: the technological shock that AI can quickly replace intellectual work is colliding with the social fear that the rules of fair competition are breaking down.
2 minutes 44 seconds for three subjects
The news agency Newsis gave the exam papers to ChatGPT and Gemini. Both systems made no mistakes on the Korean and math sections. In English, they answered all 28 reading questions correctly. The 17 listening questions weren't tested.
Here's how long ChatGPT took, according to Newsis:
| Subject | AI time | Time allowed for humans |
|---|---|---|
| Korean | 44 seconds | 80 minutes |
| Math | 1 minute 25 seconds | 100 minutes |
| English (reading only) | 35 seconds | 70 minutes (incl. listening) |
| Total | 2 minutes 44 seconds |
The source itself flags an important caveat: the listening section was left out, and both solving time and accuracy depended on the model and its reasoning settings. So you can't say AI passed "the whole Suneung" in under three minutes. The result is still remarkable, because the Suneung isn't a test of rote calculation. The Korean section asks you to spot logical relationships and implied meanings in texts. Math requires finding the right approach from a set of conditions. English tests how well you follow connections between sentences.
The leaderboard: lots of models near perfect
A more systematic look comes from the "Suneung LLM solving dashboard" built by Koo Yu-gyeom, a software engineering student at Soonchunhyang University. He has Korean and international large language models (LLMs, AI systems trained on huge amounts of text) sit the exam and compares their scores.
| Model | Provider | Score (out of 450) |
|---|---|---|
| GPT-6.2 Astra | OpenAI | 450 |
| Gemini 4 Neon | 450 | |
| GPT-6.2 Sol | OpenAI | 449.6 |
| Claude Fable 5.2 | Anthropic | 449.3 |
| Claude Opus 5.5 | Anthropic | 449.3 |
| Kimi K3.1 Pro | Moonshot (China) | 438 |
| Qwen4 Max | Alibaba (China) | 434.5 |
| DeepSeek V4.1 Flash | DeepSeek (China) | 432.7 |
| K-Exaone | LG AI Research (Korea) | 414.7 |
According to Topstar News, the K-Exaone figure comes from a report in the newspaper Edaily. The 450 points break down into 100 each for Korean, math and English, 50 for Korean history and 100 converted points for two electives in the natural or social sciences. For the electives, the dashboard averages scores across several subjects, while a real test-taker only sits one combination. So these numbers are a benchmark (a standardized performance comparison), not actual exam results.
On methodology: the dashboard uses the providers' official APIs, with no web search, calculator or other tools. Text, figures and charts were fed in, and the answers were checked against the answer key. That measures the models' raw problem-solving ability. If you type the questions into the regular chat app, different settings mean you won't necessarily get the same results. The most telling part is that several independent models from different companies all landed at the top at once. This isn't one model's lucky run. The whole leading pack has reached this level.
98 cheating cases on exam day
The Ministry of Education's figures show this isn't a theoretical problem. On exam day, 98 people were caught cheating:
- 49 were caught via an emergency alert text message,
- 44 had banned electronic devices on them,
- 5 actually used smart devices during the exam: one pair of smart glasses, one hidden camera with wireless earphones, one case of photographing the exam paper with a smartphone, and two cases of using a phone in the restroom.
In none of the five cases is there evidence yet that an AI actually supplied the answers. Owning a device and cheating with AI aren't the same thing. The government had stepped up its defenses ahead of time. Test-takers couldn't bring any writing tools except black pencils, and mechanical pencils, erasers and correction tape were handed out on site. Stickers were on hand to cover buttons in case they hid tiny cameras, and proctors were trained to recognize smart glasses and miniature cameras.
Topstar News describes the core problem like this: in the past, cheating meant knowing the answers in advance or having an accomplice. Generative AI only needs the question and produces the answer itself. If someone manages to send the questions out and get the answers back, they can score highly without solving anything themselves. Photographing, transmitting and receiving all take time, and signal jammers, bag checks and stricter supervision help. But nobody can guarantee every form of cheating will be caught. The real question is how many cases go undetected.
Other exams hit too, and a death
The problem goes beyond the Suneung. Recently, the top scorer on Korea's certified tax accountant exam admitted to cheating with a hidden camera and generative AI.
The case that shook the country most is that of Lee Jin-woo, covered in depth by the Korea Herald. The 32-year-old was a first-year dental student at Seoul National University and had scored a perfect mark on the LEET, the law school admission test. He made his story famous in the bestseller "Brain Accel," published in September: a high school dropout with weak Suneung scores who worked his way up to top test-taker. He then became a sought-after speaker. He credited his success to a digital detox, exercise and cutting out sugar.
The suspicion started with a video. On October 6, Lee showed his LEET test booklet on an educational YouTube channel. Viewers noticed it had almost no underlining or notes. More importantly, he had changed his answers to questions 25 through 29, and his original marks matched the correct answers to questions 24 through 28 exactly. That fueled speculation that he had copied answers sent from outside, possibly by an AI, and accidentally shifted them by one line. The Korea Herald stresses that the marks alone don't prove cheating. Later, inconsistencies in transcripts he had published and reports of prior criminal convictions surfaced. A lawyer filed a complaint alleging AI cheating on the Suneung and the LEET, and a second complaint alleged he had deceived the people who bought his book. Police secured his exam records.
Lee had announced he would refute the allegations in a livestream, solving the November 19 Suneung questions in public. He never showed up. He was found dead the same day, and police believe it was suicide. According to the Korea Herald, he left a note saying he was physically and mentally exhausted. It neither confirmed nor denied the allegations. Because South Korea doesn't pursue criminal cases against the deceased, the investigation is likely to be closed. Whether he actually cheated will probably never be settled in court.
If you're struggling, please reach out to a local crisis line or emergency services. In Germany, Telefonseelsorge is available free of charge around the clock at 0800 111 0 111.
Would redesigning exams help?
Korea is now debating whether to replace multiple-choice questions with essay and written-justification questions, so that reasoning gets graded rather than just final answers. Topstar News pushes back: AI also writes coherent arguments and mathematical proofs. Teachers are skeptical too. In a teachers' union survey of 1,346 high school teachers from 16 regions, 70.4 percent opposed essay formats on the Suneung. Of those, 94.3 percent feared it would further inflate the already enormous private tutoring industry, and 93.0 percent worried about fair grading and widening educational inequality.
The report lists alternatives such as oral exams, practical exams or step-by-step assessment in a controlled environment. But these are expensive, hard to grade consistently and nearly impossible to scale to hundreds of thousands of test-takers. The ministry plans to keep investigating suspicious cases, and an exam committee will decide on sanctions.
Our take
Top models acing standardized tests isn't new. What's new is how completely and how quickly they do it on one of the world's toughest school-leaving exams, using questions published that same day. That rules out the models having seen the answers during training. Treat the dashboard's scoring with caution, though. It uses averages, leaves out the listening section, and its results haven't been independently verified.
The bigger story is about society. An exam system built on the assumption that supervision prevents cheating now faces a technology that delivers answers in seconds and works with devices hidden in glasses or buttons. With a limited number of university places, every cheater who slips through takes a spot from an honest student. There's no evidence yet that this is happening on any significant scale. But precisely because it's so hard to prove, trust in exams is starting to crack. For countries with centralized exams, including in Europe, Korea is an early warning.
Sources
- Naver News (KR) topstarnews.net
- koreaherald.com koreaherald.com
- NewsData (south korea) donga.com
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.