For many years, artificial intelligence models have made a strong impression with their language processing capabilities, but have shown weakness when faced with rigorous financial tests.
However, the picture is changing rapidly. A new study shows that the top 6 AI models today have passed all 3 levels of the Financial Analyst (CFA) exam, which is one of the most difficult certificates in the investment field.
CFA is considered a golden ticket for financial experts, opening up global career opportunities and high income.
The exam consists of 3 levels with increasing difficulty, requiring not only knowledge but also the ability to apply, synthesize and judge. AI's passing of the entire program shows a significant step forward in reasoning capabilities.
According to the study, Google's Gemini 3.0 Pro model reached a record score of 97.6% at Level I. GPT-5 led at Level II with 94.3%, while Gemini continued to show its advantage at Level III, which was the most difficult part with a rate of exactly 92%.
Other models such as Claude Opus 4.1, Grok 4 and DeepSeek-V3.1 also achieved the same results. In total, these systems were tested on over 980 simulation questions, based on official documents and the prestigious test of the CFA Institute.
The significance of this result lies in the structure of the CFA exam. Level I tests basic knowledge such as economics, financial reporting, ethics and qualitative methods. Level II requires the application of knowledge in specific tinh huong studies.
Level III requires candidates to synthesize information, build investment strategies and explain arguments in writing. These are skills that AI has often struggled with in the past.
Researchers say the difference lies in the architecture of the new models. Instead of just remembering and tracing information, the new generation of AI has the ability to reason in many steps, track long-term context and apply flexible rules according to conditions.
In particular, the capacity for quantum reasoning has been significantly improved, when the error rate in toanistic sections is almost zero.
However, AI is not yet perfect. Professional ethics continues to be a weakness, as models sometimes apply rules in a subtle way.
The essay answer section is also controversial, because scoring with other AI can create a deviation from the strict assessment of human judges.
More importantly, passing the exam does not mean being able to replace CFA experts. Portfolio management, customer communication, market fluctuations handling and legal responsibility are still areas that require experience and human judgment.
However, this achievement shows that AI has surpassed a new threshold, capable of handling many complex analysis tasks.
For the financial industry, AI is both a strong support and a big challenge. When machines can pass professional exams, the value of these certificates can shift from testing knowledge to assessing decision-making ability, ethics and responsibility in the real world.
