Riffing on Aya's sli.do survey from yesterday's The New Investment Order conference, "AI that writes better reports than your analyst"... AI can pass CFA III (!)
https://www.cnbc.com/2025/09/24/ai-cfa-exam-pass-minutes-study.html
(I'm glad we'll still have jobs for work that requires judgment and navigating human-generated complexity.)
Recap from The Rundown newsletter: Research from NYU has found that frontier models from OpenAI, Google, and Anthropic can now pass all three levels of the CFA (chartered financial analyst) exam, including difficult Level III essay questions that eluded them two years ago.
The details: NYU Stern and GoodFin researchers tested 23 language models on mock CFA Level III exams, finding nine models achieved passing scores above 63%.
OpenAI's o4-mini scored highest at 79.1% on the challenging essay portion, with Gemini 2.5 Pro and Claude 4 Opus reaching 75.9% and 74.9%.
Models completed the exam in minutes versus the 1,000 hours humans typically spend studying across multiple years for all three levels.
Human graders also consistently scored AI essay responses 5.6 points higher than automated grading systems.
Why it matters: The leap from failing essay sections two years ago shows the huge shift in analytical capabilities, with reasoning models perfectly suited for the complex thinking process. With AI's rise, human aspects like client relationships and contextual judgement will become bigger factors than research reports and investment rationales.
------------------------------
Kara K.W. Byun
Head of Fintech
------------------------------