Artificial intelligence has reached a major milestone in the finance world: advanced AI models can now pass the notoriously difficult Level III CFA exam—an achievement that typically requires human candidates years of experience and over 1,000 hours of study.
In a new study from NYU’s Stern School of Business and AI wealth management platform GoodFin, 23 large language models were tested on mock versions of the CFA Level III exam. Known for its emphasis on portfolio management and wealth planning, this final level of the CFA charter is especially challenging due to its complex essay questions.
Historically, these essay sections have been a major stumbling block for AI systems. But recent advancements have changed the game. Leading models like OpenAI’s o4-mini, Google’s Gemini 2.5 Pro, and Anthropic’s Claude Opus performed exceptionally well—without any domain-specific training. Instead, they leveraged a technique called “chain-of-thought prompting,” which allows them to break down problems step by step—an ability earlier AIs lacked.
Just two years ago, AI systems could handle the multiple-choice-heavy Levels I and II but consistently failed to grasp the nuance of Level III. The latest findings suggest that gap is closing fast.
The results raise big questions about the future of financial advising. “There’s absolutely a future where this technology transforms the industry,” said GoodFin CEO Anna Joo Fee, according to CNBC. However, she stopped short of declaring the end of human CFAs. “There are things like context and intent that are still difficult for machines to understand,” she noted.
Interestingly, the most accurate models also came with steep costs—some requiring up to 11 times more computing power than their peers, according to the press release.