12月26日，"2025金融大模型评测体系暨金融评测基准"在上海市委金融办、上海市委网信办、上海市经信委及上海徐汇区人民政府指导、由上海人工智能实 验室与上海库帕思科技有限公司联合主办的发布会上正式落地。 记者现场了解到，当下，金融大模型仍存在着实时性与动态适应性不足、偏见和公平性、数据安全与隐私、领域知识深度不足及透明度和可解释性薄弱等相 对局限性。 针对以上痛点，该项最新评测体系汇聚了4个公开数据集与22个自建数据集，约3.6万条评测数据，坚持科学与鲁棒的评测过程，采用循环选项打乱机制和多 样化提示词，并研发金融裁判大模型，实现评测全流程自动化、标准化，为上海金融领域银行、券商、基金、投资等企业或者机构人员提供权威、精准的大 模型能力评估，助力机构选型、优化及风险把控。 21世纪经济报道记者 余纪昕 上海报道 该场发布会上，上海市委金融办副主任、一级巡视员葛平指出，人工智能正在深刻重塑金融行业发展格局，大模型技术在应用场景深化、关键要素强化和应 用生态协同推进三方面加速赋能。 他强调，去年由库帕思、上海人工智能实验室等机构发布的全国首个"以金融业务为中心"的金融大模型评测体系，为行业提供了科学选型与能力对 ...

Core Viewpoint - The "2025 Financial Large Model Evaluation System" was launched in Shanghai, marking a significant step in the intelligent transformation of the financial industry, aiming for higher quality and more reliable applications of AI technology in finance [1][2]. Group 1: Evaluation System Overview - The evaluation system is a collaborative effort between Shanghai Artificial Intelligence Laboratory and KuPass Technology, showcasing technological achievements in financial model assessment [1]. - The system is designed to provide a scientific benchmark for financial institutions, facilitating the selection and capability comparison of large models [1][2]. - The comprehensive upgrade of the evaluation system aims to support Shanghai's goal of becoming a globally influential financial technology center [1]. Group 2: Data and Methodology - The evaluation system integrates 4 public datasets and 22 self-built datasets, totaling approximately 36,000 evaluation data points [2]. - It employs a robust evaluation process with mechanisms like randomized options and diverse prompts, alongside the development of a financial referee large model for automated and standardized evaluation [2]. - The system aims to assist banks, brokerages, funds, and investment institutions in accurately assessing large model capabilities, optimizing selections, and managing risks [2]. Group 3: Reports and Applications - A joint report titled "Financial Large Model Application Evaluation Report (2025)" and a dataset titled "Financial Large Model Evaluation Dataset (2025)" were also released, focusing on real financial business scenarios [2]. - The report explores new concepts, mechanisms, and methods for applying large models in vertical financial fields, supporting institutions in scientific selection and cost reduction [2]. - The initiative is expected to accelerate the large model's implementation in key areas such as investment research, risk control, and customer service [2].