OpenAI’s latest AI model, GPT‑6 Astra, has become the first system to score a perfect 450 on South Korea’s College Scholastic Ability Test (CSAT), marking a new milestone in the model’s demonstrated reasoning and subject mastery.
Perfect CSAT score
According to results published on GitHub on Sunday, GPT‑6 Astra was the only evaluated large language model to achieve a flawless 450 in the 2026 CSAT LLM Solution Log, an offline, no‑internet evaluation covering Korean language, English, mathematics, Korean history and four elective subjects: Physics I, Chemistry I, Life Science I and Society and Culture. The test is widely regarded as one of the most demanding standardized exams in the world, making the perfect score a high‑profile signal of the model’s cross‑domain competence.
Broader benchmark performance
The CSAT result sits alongside other top scores reported for Astra since its rollout began on September 3, 2026. OpenAI and independent trackers have highlighted Astra’s 99.9% on the ARC‑AGI‑3 generalisation benchmark, 98% on FrontierMath Tier 4, and a perfect 100% on ExploitBench, a security‑focused evaluation. Company materials also show Astra leading or matching rivals on computer‑use and coding tasks, including strong results on OSWorld 2.0 and ScreenSpot‑Pro.
Limited rollout and safety guardrails
Astra is being released gradually, initially to a limited set of organisations in OpenAI’s Daybreak Access programme, with wider availability planned for ChatGPT Plus, Pro, Business and Enterprise users, as well as via the OpenAI API and selected cloud platforms. The launch comes with heightened safety measures: OpenAI rates Astra as “Critical” under its Preparedness Framework for cybersecurity capability, meaning it can, in principle, find and exploit novel vulnerabilities with limited human guidance, prompting additional training and deployment restrictions to reduce misuse risks.
Market and policy ripple effects
The model’s capabilities have already influenced markets and policy debates. South Korean semiconductor stocks, including Samsung and SK Hynix, rose sharply after the Astra announcement, reflecting expectations of higher AI compute demand. At the same time, regulators and security researchers are scrutinising the implications of “Critical”‑rated models that can operate computers autonomously and potentially evade human monitoring, issues OpenAI acknowledges in its safety overview.
