More and more exams are falling to the rapid progress of AI.
AI can now pass the CFA (Chartered Financial Analyst) Level 3 exams, a study by New York University Stern School of Business and Goodfin, an AI-powered wealth-management platform, has found. Thus far, AI had been able to pass the CFA Level 1 and Level 2 exams, which are simpler and only have multiple choice questions. But AI can now pass the Level 3 CFA exam as well, which includes both multiple choice and essay questions.
The CFA is one of the most prestigious certifications in finance. Candidates must pass three exams, Level 1, Level 2 and Level 3 to be able to be awarded the certification. An exam can be attempted only if a candidate is able to clear the previous levels. It typically requires 1,000 hours of effort to pass all three levels.
AI systems can now pass Level 3 exam in minutes. As per the leaderboard, Gemini 2.5 Pro scored the highest overall score including the multiple choice and essay questions, followed by o4-mini and Claude Opus 4. In the Multiple Questions section, Gemini 2.5 Pro scored of 77%, followed by o4-mini which scored 68% and Claude Opus 4, which scored 60%. The best-performing non-reasoning model was Grok 3, which had an overall score of 2.02 and got 65% of multiple choice questions correct.
The CFA is the latest exam to fall in the ongoing AI onslaught. Earlier this year, Google Gemini 2.5 had performed better than the best-performing human student at the IIT-JEE Advanced 2025. Apart from such exams, AI models are doing exceedingly well at math and programming competitions. Both Google and OpenAI have delivered gold-medal performances at the International Math Olympiad, and more recently, both labs also delivered gold-medal performances at the International Collegiate Programming Contest 2025, with OpenAI performing better than all human teams, while Google performed better than all human teams but one. And with this progress only seeming to accelerate, it appears that humans need to quickly prepare for a world where AI systems are better than them at the most-highly-paid knowledge work.