In minutes, AI passes the most difficult CFA exam

According to NYU research, frontier models from Google, Anthropic, and OpenAI can now successfully complete all three levels of the CFA (chartered financial analyst) test, including challenging Level III essay problems that they were unable to complete two years ago. The specifics: Researchers from NYU Stern and GoodFin examined 23 language models on simulated CFA Level III exams and discovered that nine of them passed with passing percentages higher than 63%. On the difficult essay section, OpenAI's o4-mini received the highest score of 79.1%, followed by Gemini 2.5 Pro and Claude 4 Opus, which earned 75.9% and 74.9%, respectively. In contrast to the 1,000 hours that humans usually spend studying for all three levels over several years, models finished the exam in a matter of minutes. Additionally, AI essay replies were regularly rated 5.6 points higher by human graders than by automatic grading methods. With reasoning models that are ideal for the intricate thought process, the leap from failing essay portions two years ago demonstrates the significant shift in analytical capacities. As AI advances, human elements such as client connections and contextual judgment will take precedence over research findings and investment justifications.