Kevin Cartr / iStock / Getty Images Plus
Follow ZDNET: Add us as a preferred source on Google.
ZDNET's key takeaways
Frontier models passed the CFA Level III exam.
Less than half of (human) candidates passed the exam in February.
AI is rapidly becoming better at certain tasks.
Some tasks that demand Herculean cognitive effort from humans are trivially easy for AI systems, which are designed to detect and replicate complex patterns gleaned from enormous troves of data. The technology has already scored top marks in world-class competitions in mathematics and coding; soon -- according to the vision of some developers -- help human researchers make new scientific discoveries.
Also: OpenAI tested GPT-5, Claude, and Gemini on real-world tasks - the results were surprising
Now, AI's capabilities are also rapidly catching up with the most skilled human financial analysts.
A new study conducted by the New York University Stern School of Business and AI-powered wealth management platform GoodFin found that some frontier AI models were able to successfully pass a mock version of the Chartered Financial Analyst (CFA) Level III exam, widely regarded as the world's most difficult and prestigious testing benchmark for aspiring investment management professionals.
Passing the test
The study analyzed the capabilities of 23 industry-leading models, both proprietary and open-source, from developers like Google, OpenAI, Anthropic, Meta, xAI, and DeepSeek.
Also: AI helps strong dev teams and hurts weak ones, according to Google's 2025 DORA report
Previous studies had demonstrated that AI could pass the CFA Levels I and II exams, but it struggled with the third and final (and most difficult) phase. The Level III exam, designed to test candidates' ability to apply their knowledge of portfolio management and wealth planning to hypothetical real-world scenarios, includes a set of multiple-choice as well as essay questions.
"This dual format comprehensively tests higher-order cognitive skills including analysis, synthesis, and professional judgment over rote memorization," the researchers behind the new study wrote in a paper that was originally posted to the preprint server site arXiv in June. "The exam's rigorous standards...make it an excellent benchmark for assessing advanced financial reasoning capabilities."
Less than half (49%) of human test-takers passed the Level III exam in February, according to the CFA Institute.
The results
A handful of reasoning models, which specialize in breaking problems down into a series of sub-problems that are then tackled successively, were able to pass the mock exam.
Also: This app will pay you $30/day to record your phone calls for AI - but is it worth it?
OpenAI's o4-mini scored the top spot with an overall score of 79.1%, followed by Google's Gemini 2.5 Flash, which scored 77.3%. The passing threshold for the exam is 63%.
The researchers note in their paper that while most of the models included in the study scored in the same range on the multiple-choice segment of the exam (around 71%-75%), their scores varied much more widely on the more rigorous and challenging essay segment.
"This suggests that simpler, more straightforward tasks have become commoditized across models, whereas complex and nuanced reasoning tasks still differentiate frontier and reasoning-enhanced models from their peers," they write.
Also: How AI can help you manage your finances (and what to watch out for)
A recent report from Microsoft listed personal financial advisors as one of the forty job categories most likely to be taken over by AI. Still, Anna Joo Fee, founder and CEO of GoodFin, told CNBC that she's not concerned about immediate replacement.
"There are things like context and intent that are hard for the machine to assess right now," she said. "That's where a human shines, in understanding your body language and cues."