GPT-6 Astra aces Korean college entrance exam
A student fills out an OMR answer sheet during a September mock test for the 2027 College Scholastic Ability Test at Seoul Girls' High School in Mapo District, Seoul, Wednesday. Yonhap
New system used 357,000 tokens on 2026 test, beating rivals with design that reuses earlier reasoning
An analysis showed OpenAI's new artificial intelligence (AI) model earned a perfect score on the College Scholastic Ability Test (CSAT) for the first time. While top-tier AI models have previously posted near-perfect scores, this result is drawing attention because it used fewer tokens than existing models.
According to results posted on GitHub on Sunday, OpenAI's new GPT-6 Astra was the only evaluated model to score a perfect 450 in the 2026 CSAT LLM Solution Log. The evaluation tested models on questions from the 2026 CSAT across Korean language, English, mathematics, Korean history and four elective subjects — Physics I, Chemistry I, Life Science I and Society and Culture — without internet access.
Astra was the first AI model to achieve a perfect score across all tested subjects in the evaluation. In February, Google's Gemini 3.1 Pro earned a perfect score taking only two electives — Chemistry I and Life Science I — but missed questions in Physics I and a social studies subject under this broader format. In the rankings released Sunday, OpenAI's GPT-5.6 placed second with 448.5 points, while GPT-5.4 placed third with 448. Anthropic's Claude Fable 5.1 placed fourth with 447.5 and Gemini 3.1 Pro placed fifth with 445.
GPT-6 Astra used 357,000 tokens, the lowest reported total among publicly available models. GPT-5.6 used 429,000 tokens and Claude Fable 5.1 used 562,000. Because the exam questions and answers were publicly available, the result cannot rule out prior exposure through training data, though competitor models like Claude Fable 5.1 and Gemini 3.1 Pro faced the same conditions.
An image generated by artificial intelligence
The model's high performance and lower token usage is attributed to a reasoning-linked design that retains context during repetitive tasks. A provider adapter harness compresses and reuses information from earlier reasoning steps, allowing the AI to take the test while mimicking the approach of a top-scoring student. Lee Seung-hyun, an adjunct professor at Hanyang University, said the key is not simply that the model became smarter, but that it can retain prior reasoning steps, compress context and revise its plan.
"It means that rather than thinking harder each time, it continued with what it had already figured out before," Lee said.
OpenAI described the result as encouraging, noting that the model achieved efficient processing using hyperscale AI computing infrastructure. An OpenAI official said the milestone demonstrates that models can become more powerful while improving performance and lowering costs.
Industry experts cautioned, however, that model potential should be evaluated by its ability to solve open-ended problems rather than standardized exam scores alone. A Korean AI industry insider said it will be more meaningful if AI tackles real-world workplace challenges where there are no predetermined answers.
This article from the Hankook Ilbo, the sister publication of The Korea Times, is translated by a generative AI system and edited by The Korea Times.