AI Aces the CFA Level III Exam: The End of Financial Advisory as We Know It?

The moment came quietly, without fanfare or press conferences. In research labs at NYU Stern and GoodFin, artificial intelligence models started passing the most grueling exam in finance — the CFA Level III — in minutes rather than the years humans typically need.

The results weren’t close calls either. OpenAI’s o4-mini scored 79.1%. Google’s Gemini 2.5 Pro hit 75.9%. Anthropic’s Claude Opus 4 achieved 74.9%. All sailed past the 63% threshold that stumps most human candidates.

“This changes everything,” one wealth management executive told us. (We promised to keep their identity anonymous in exchange for their insights.) “If AI can pass the hardest financial exam, what exactly are we paying human analysts for?”

The Breakthrough That Redefined the Game

The researchers tested 23 models across three different approaches to problem-solving. Zero-shot prompting, asking directly for answers, yielded decent results. But the real magic happened with “chain-of-thought prompting,” where AI explains its reasoning step-by-step before reaching conclusions.

The pattern was stark: straightforward multiple-choice questions had become commoditized across AI platforms. All the top models scored around…