The three-part test to become a chartered financial analyst is widely viewed as one of the hardest exams in finance. But new research shows that many of today’s AI models can pass it in a matter of minutes—and they continue to improve at a rapid pace.
In the study, researchers from GoodFin, an AI-powered wealth advisory firm, and the NYU Stern School of Business pitted 23 AI models against mock versions of the CFA Level III exam, including its notorious essay section. Responses were then scored by human graders and LLMs.
The researchers found that many cutting-edge models comfortably achieved a passing score of 65% or above. OpenAI’s o4-mini model scored 79.1%, Google’s Gemini 2.5 Flash scored 77.3%, and Anthropic’s Claude Opus 4 scored 74.9%. Models developed by DeepSeek and xAI also earned their CFAs.
Read more | BARRON’S