大模型财经写作大比拼：谁才是上市公司研报的“笔杆子”冠军？

Core Insights - The demand for high-quality financial analysis is increasing due to the explosion of financial information, and traditional manual analysis is struggling to keep pace with rapid market changes [1] - The rise of large model technology offers new solutions to these challenges, although its application in financial analysis faces hurdles such as data processing accuracy and understanding market sentiment [1] Evaluation of Large Models - Five representative large models were evaluated: GPT-4, Claude, Gemini, Wenxin Yiyan, and Tongyi Qianwen, tasked with writing a financial analysis of Amazon's Q2 2025 earnings report [1] - The evaluation criteria included data accuracy, richness, writing ability, logical reasoning, innovation, and writing speed, revealing subtle differences in performance among the models [1] Performance Highlights - GPT-4 excelled in data accuracy and logical reasoning, with its article fully aligning with the provided financial data and demonstrating high data processing and expression capabilities [2] - Claude and Gemini followed closely, with Claude showcasing professional language and clear logic, while Gemini highlighted the contrast between strong earnings and stock price decline, reflecting market complexities [4] - Wenxin Yiyan and Tongyi Qianwen also displayed notable strengths, with Wenxin Yiyan adopting a structured format similar to professional research reports and Tongyi Qianwen innovatively categorizing investment advice into short, medium, and long-term [4] Limitations of Large Models - The evaluation revealed limitations in large models' ability to delve deeper into economic implications and industry trends behind the data, despite accurate citation of core figures [6] - Articles generated by large models tended to lack groundbreaking insights and forward-looking judgments, indicating a conservative approach to innovative viewpoints [6]