Internet

“AI Took the College Scholastic Ability Test, Too”… OpenAI and Google Scored a Perfect 450—How Did South Korea’s AI Perform?

Koo Yu-gyeom of Soonchunhyang University Unveils AI-Powered College Entrance Exam Score Dashboard OpenAI and Google Tie for First Place… Anthropic and Chinese AI Companies Show Strong Performance Domestic AI: LG Corp. Leads with 414.7 Points Upstage, KRAFTON, and KTCorporation, in that order 2027 College Scholastic Ability Test to Be Administered Without Calculators or Search Functions

Kim Hyun-ah
2026-11-20 06:02:22
[Edaily Reporter Kim Hyun-ah ] If artificial intelligence (AI) had taken this year’s College Scholastic Ability Test (CSAT), what score would it have received? An online dashboard that allows users to compare the CSAT scores of major domestic and international AI models at a glance has been released. AI models from OpenAI and Google both scored a perfect 450 points, while among domestic models, “K-ExaOne” from the AI Research Institute at LG Corp.(003550) received the highest score.

According to the “2027 CSAT Large Language Model (LLM) Solution Dashboard,” developed on the 20th by Koo Yu-gyeom of the Department of Computer and Software Engineering at Soonchunhyang University, OpenAI’s “GPT-6.2 Astra” and Google’s “Gemini 4 Neon” tied for first place with a perfect score of 450 points.

On the 19th, the day before, as soon as the CSAT questions and answers were released, Mr. Koo had domestic and international AI models solve the problems and reflected the results on the dashboard. Users can compare scores by subject—including Korean, Math, English, Korean History, and Inquiry—as well as the score differences between models. The 450-point total is the sum of the raw scores from the five required CSAT sections (Korean, Math, English, Korean History, and two Inquiry subjects).

Written using generative AI. Source: Mr. Koo Yu-gyeom, Department of Computer and Software Engineering, Soonchunhyang University

OpenAI and Google Score Perfect Marks… Chinese AI Also
Scores
in the 430s
While OpenAI and Google tied for first place, other U.S. AI models also achieved high scores. OpenAI’s “GPT-6.2 Sol” ranked third with 449.6 points, while Anthropic’s “Claude Fable 5.2” and “Claude Opus 5.5” tied for fourth place with 449.3 points each. OpenAI’s “GPT-6.1 Terra” scored 445.8 points.

Chinese AI companies also made it into the top ranks. MoonShot AI’s “Kimi K1 Pro” scored 438 points, Alibaba’s “Qwen 4 Max” scored 434.5 points, and DeepSeek’s “DeepSeek V4 Flash” scored 432.7 points. While U.S. models dominated the top ranks, Chinese models also achieved scores in the 430s.

LG Corp. Leads South Korean AI… Followed by Upstage, KRAFTON, and KTCorporation
Among domestic AI models, LG AI Research’s “K-EXAONE 2.5” received the highest score of 414.7 points. It ranked 12th among the major domestic and international models that participated in the test.

Upstage’s “Solar Pro 4” ranked 14th with 392.8 points, KRAFTON(259960)’s “Raon-1-mini-preview” ranked 15th with 357.2 points, and KTCorporation(030200)’s “Mi:dm K 3.0 Pro” ranked 16th with 352.8 points.

Even K-ExaOne 2.5, the top-ranked model in Korea, trailed the top-scoring model by 35.3 points.

Each model has different development objectives and scales. KRAFTON’s Raon-1 Mini Preview focuses on Korean language processing and Agent AI tasks. KTCorporation’s Mi:dm K 3.0 Pro is a model designed to enhance Korean language reasoning capabilities. Both models apply the Model of Experts (MoE) approach, which selectively activates only the necessary specialized neural networks to reduce computational load.
Koo Yu-gyeom: “Evaluation Immediately After Questions Were Released… Reduces the Possibility of Memorizing Correct Answers”
Koo
Yu-gyeom
, who developed this dashboard, goes by the nickname “hehee9” and has been publishing AI CSAT solution results since last year. By using the familiar CSAT exam, he has made it easy for the general public to compare the problem-solving abilities of different AI models.

This evaluation covered all elective subjects in Korean and Math, as well as all subjects in English, Korean History, and Science and Social Studies.

Mr. Koo explained, “Since we solve the problems almost as soon as they are released, we can significantly reduce the likelihood that the model has already memorized the correct answers.”

The evaluation utilized the AI models’ application programming interfaces (APIs) directly. External tools—such as search engines or calculators—available on general web services like ChatGPT could influence the problem-solving process, so these were excluded. Mr. Koo focused on comparing the models’ inherent problem-solving capabilities without relying on external tools.

Exam questions were provided in text format, and tables, underlines, and other elements were reproduced using Markdown. Figures and graphs were attached as high-resolution images. Models that do not support image input were instructed to solve the problems without the figures.

Scores can be viewed as the sum of raw scores by subject and as a converted score out of 450. Mr. Koo explained, “Since the weighting of common subjects, elective subjects, and the inquiry section varies, even a single point can have a different impact on the final converted score depending on the subject.”

A feature was also added to distinguish between AI models available for free and those requiring a paid subscription. This allows users to compare not only test scores but also performance differences based on usage costs.

These results demonstrate the level to which AI’s knowledge comprehension and problem-solving abilities have advanced. However, since each model has a different development purpose and test conditions—such as whether image input is supported—vary, it is difficult to judge the AI’s overall competitiveness based solely on the published scores.

The dashboard operated by Mr. Koo can be viewed at the “2027 College Scholastic Ability Test LLM Solution Dashboard.”

Economy

Corporation

IT·Science

Economy

From Hanjin to Doosan… Credit Rating Upgrades for BBB-Rated Companies Signal a ‘Positive Trend’

As interest rate hikes increase the burden of corporate financing, attention is focusing on the possibility of credit rating upgrades for BBB-rated companies. With BBB-rated companies such as Hanjin a…
2026-11-19 19:02:04

IT·Science

“AI Took the College Scholastic Ability Test, Too”… OpenAI and Google Scored a Perfect 450—How Did South Korea’s AI Perform?

If artificial intelligence (AI) had taken this year’s College Scholastic Ability Test (CSAT), what score would it have received? An online dashboard that allows users to compare the CSAT scores of maj…
2026-11-20 06:02:22