Artificial intelligence has hit a new milestone in the world of finance: Top models can now breeze through the notoriously tough Level III CFA exam in minutes, a task that typically demands years of preparation and roughly 1,000 hours of study from human candidates. New research by New York University's Stern School of Business and the AI wealth-management platform GoodFin put 23 large language models to the test on mock versions of the CFA Level III exam, which is known for its focus on portfolio management and wealth planning, per CNBC.
The challenge isn't just multiple choice: The exam's essay questions have long stumped even the most advanced AI. But the latest models, including OpenAI's o4-mini, Google's Gemini 2.5 Pro, and Anthropic's Claude Opus, cracked the code without any domain-specific training, per a press release, using what's known as "chain-of-thought prompting." This approach lets them explain step-by-step reasoning—something earlier AIs couldn't handle. Just two years ago, AI could squeak by Levels I and II of the exam but would hit a wall with Level III.
The rapid leap in technology prompted researchers to revisit the test, aiming to see if machines could manage the kind of high-stakes, nuanced thinking demanded in real-world finance. GoodFin founder and CEO Anna Joo Fee sees big changes on the horizon. "There's absolutely a future where this technology transforms the industry," she says, per CNBC. Still, Fee isn't predicting the demise of human CFAs just yet. "There are things like context and intent that are hard for the machine to assess right now," she says. The press release also noted that "models with the highest accuracy also had the highest compute costs—up to 11x more."
,
uBlock 
,
and
Ghostery 
)
