Reasoning Models Ace the CFA Exams

2025-12-09

AI Summary

Prior large language models (LLMs) have underperformed on high-stakes, domain-specific professional certification exams—particularly the Chartered Financial Analyst (CFA) program—due to challenges in rigorous financial reasoning and structured problem-solving. Method: This study conducts a systematic evaluation of state-of-the-art reasoning models—including Gemini 3.0 Pro, GPT-5, Claude Opus, Grok, and DeepSeek—on full-scale CFA Level I–III mock examinations comprising 980 expert-curated finance questions. To address reasoning limitations, we employ multi-step chain-of-thought prompting and structured answer generation. Contribution/Results: We provide the first empirical evidence that most evaluated models achieve passing scores across all three CFA levels. Notably, Gemini 3.0 Pro attains 97.6% accuracy on Level I; GPT-5 and Gemini variants significantly surpass historical human average scores on Levels II and III. These findings demonstrate that next-generation reasoning models possess robust capabilities for complex financial analysis, offering foundational empirical validation and a methodological framework for AI-augmented professional credentialing.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Previous research has reported that large language models (LLMs) demonstrate poor performance on the Chartered Financial Analyst (CFA) exams. However, recent reasoning models have achieved strong results on graduate-level academic and professional examinations across various disciplines. In this paper, we evaluate state-of-the-art reasoning models on a set of mock CFA exams consisting of 980 questions across three Level I exams, two Level II exams, and three Level III exams. Using the same pass/fail criteria from prior studies, we find that most models clear all three levels. The models that pass, ordered by overall performance, are Gemini 3.0 Pro, Gemini 2.5 Pro, GPT-5, Grok 4, Claude Opus 4.1, and DeepSeek-V3.1. Specifically, Gemini 3.0 Pro achieves a record score of 97.6% on Level I. Performance is also strong on Level II, led by GPT-5 at 94.3%. On Level III, Gemini 2.5 Pro attains the highest score with 86.4% on multiple-choice questions while Gemini 3.0 Pro achieves 92.0% on constructed-response questions.

Problem

• Evaluating reasoning models' performance on CFA mock exams
• Assessing if advanced AI can pass all three CFA levels
• Comparing top models' scores across different exam formats

Innovation

• Advanced reasoning models achieve high CFA exam scores
• Models like Gemini and GPT-5 outperform previous LLMs
• Performance is evaluated across multiple exam levels and formats