多元金融行业研究专题报告：同花顺vs东方财富，金融大模型的体验与对比如何？

Investment Rating - The report maintains a "Positive" rating for the industry, indicating an optimistic outlook for the sector [4] Core Views - Both HithinkGPT (Tonghuashun) and Miaoxiang (Eastmoney) have launched AI-powered financial models in January 2024, showcasing strong coverage, professionalism, timeliness, and human-computer interaction capabilities [1][3] - HithinkGPT demonstrates stronger framework and timeliness, while Miaoxiang excels in integrating news and research reports, offering more cautious stock recommendations [1][3] - Both models use similar transformer-based decoder-only architectures, with HithinkGPT offering parameter sizes of 7B, 13B, 30B, 70B, and 130B, and Miaoxiang offering 7B, 13B, 34B, 66B, and 104B [3][6] - HithinkGPT has pre-trained financial data exceeding trillions of tokens, while Miaoxiang uses 1.2 trillion tokens, with both models adding hundreds of billions of tokens monthly [3][6] Model Comparison - HithinkGPT is integrated into Tonghuashun's mobile trading app and financial terminal iFinD, while Miaoxiang is available as a standalone app [3][7] - Both models focus on intelligent investment advisory and research, providing services such as stock analysis, news, and smart writing [3][7] - HithinkGPT offers additional features like Feiyan creation, dynamic dashboards, drawing, meeting minutes, document reading, AI image recognition, and translation [7] User Experience - Both models cover a wide range of markets including A-shares, funds, ETFs, Hong Kong stocks, US stocks, bonds, and macroeconomics [3][12] - They exhibit strong professionalism by leveraging high-quality financial data and timely market information [3][12] - However, there is room for improvement in contextual understanding, depth of analysis, fundamental analysis, and handling of overseas data [3][12] Performance Analysis - HithinkGPT provides a more structured analysis, focusing on fundamentals, technicals, capital flows, and news, while Miaoxiang integrates research reports and news, resembling a professional investment advisor [17][20] - HithinkGPT shows better timeliness in responding to new company announcements, while Miaoxiang is slower [17] - In stock recommendations, HithinkGPT offers direct investment advice, whereas Miaoxiang is more cautious [17] Limitations - Both models struggle with contextual understanding and require precise questions to avoid misinterpretation [12] - They lack depth in fundamental analysis, often focusing on technical aspects like valuation, technical analysis, and capital flows [12] - Coverage of overseas markets remains limited, with room for improvement in the quality of responses to international queries [12][16]