华南理工大学许言午教授则从跨行业经验出发，为金融AI的发展提供了另一种参照。他表示，很多人 直觉上觉得AI在金融领域"存在感不强"，其实并不准确。AI早已深度参与保险定价、资产评估和量化交 易，只是这些价值并不直接呈现在ToC产品中，因此"看不见"。 本报讯 2月5日，围绕"信贷多模态AI如何定标准"，奇富科技举办了一场聚焦产业与学术前沿的直播讨 论。直播中，奇富科技联合复旦大学、华南理工大学研究人员近期发布的首个面向信贷场景的多模态评 测基准FCMBench-V1.0成为核心议题。 复旦大学陈涛教授认为，好的数据集本身就是对"好问题"的定义。重要的是，金融AI不能停留在通用模 型的预训练与微调阶段，而应构建内生的金融思维链，让模型天然理解利率、规则与风险，实现安全可 信的推理能力。这也是学界与产业必须协同解决的问题。 （文章来源：证券日报） 在这场对话中，来自产业一线与学术前沿的三位嘉宾，从不同视角指向了同一个问题：如果没有统一标 准，金融AI很难真正落地。 作为奇富科技多模态负责人，杨叶辉首先从产业实践谈起。他用"锄头与土地"做比喻，形象阐释了AI与 应用场景之间的关系：AI是工具，而金融、医疗这样的高门 ...

2月5日，围绕"信贷多模态AI如何定标准"，奇富科技举办了一场聚焦产业与学术前沿的直播讨论。直播 中，奇富科技联合复旦大学、华南理工大学研究人员近期发布的首个面向信贷场景的多模态评测基准 FCMBench-V1.0成为核心议题。该评测基准源自真实信贷业务场景，围绕多模态感知、推理与决策等关 键环节设计评估任务，并同步开源数据集与评测工具，试图为金融AI建立一把可被广泛认可的"尺子"。 图 1："信贷多模态AI如何定标准"主题直播现场 对比医疗AI长达十余年的研发与审批周期，许言午认为，金融行业更短的业务迭代周期，反而为模型 评测和更新提供了现实土壤。他将数据集的发展分为三个阶段：先夯实数据质量，再通过学术与赛事运 营形成影响力，最终获得行业层面的官方认可，成为类似托福、雅思那样的"准入门槛"。在他看来， FCMBench正处在一个极具潜力的起点。 2：奇富科技多模态负责人杨叶辉直播分享现场 "评测先行，本质上是在做一把尺子。"杨叶辉指出，当前金融机构在选择模型和方案时，常常陷入"不 同模型分别声称得到了95分和98分，到底哪个好？"的困惑。没有统一、公平、公开的评测体系，决策 就容易失焦。FCMBench的价 ...

【环球网科技综合报道】2月5日，国内首个工业级AI漫剧智能体生产平台"纳米漫剧流水线"（简称"纳米漫剧"）正式对外 发布，并率先落地短剧产业高地郑州。在同期举行的闭门路演会上，近百家短剧从业机构和影视公司到场交流。据悉，该 产品公测期间即受到市场热捧，邀请码"一码难求"，并有多家头部公司宣布切换或接入"纳米漫剧"进行核心项目制作，引 发行业加速布局工业化漫剧生产。行业人士分析，该产品有望推动AI漫剧进入"量产精品"新阶段。 值得一提的是，这场闭门会堪称一场行业的"双向奔赴"。此前，"纳米漫剧流水线"在公测阶段已"一码难求"，而来自早期 合作伙伴的实际成果，更印证了其作为产业变革引擎的潜力：奇想文化在播放量超10亿的《全民诡异》系列后续制作中全 面采用该流水线；红猪动漫将主力工具切换为纳米，助推《丧尸之王》登上热榜第二名，并已携手筹备春节档新剧。创作 者杨涵涵使用该工具48小时打磨的《霍去病》近5分钟短片，一经发布便在各大社群和平台引发广泛传播，并在外网疯 传。这场活动不仅是一次行业的双向奔赴，更清晰地表明：一款真正解决痛点的工具，已成为头部玩家和行业新贵为赢得 漫剧商业价值的重磅押注与战略重心。市场看好，也验 ...

​Qfin Holdings, Inc. (NASDAQ:QFIN) is one of the Best Small Cap Stocks Ready to Explode in 2026. On January 28, Richard Xu from Morgan Stanley reiterated a Buy rating on the stock with a $50 price target. ​Analyst Xu of Morgan Stanley noted that he sees a mix of potential headwinds and competitive strengths for the company, with the positive factors outweighing the negative ones. Some of the headwinds identified by the analyst include new rules in microloans, which will cap yields at 12% and are expected ...

记者｜郭聪聪 编辑｜周炎炎 2025年12月31日，曾以主导滴滴与快的合并、美团与大众点评合并等重磅交易闻名的"并购之 王"华兴资本，宣布 将通过全资子公司以3.08亿元的代价，收购奇富科技旗下本息总额高达 74.29亿元的个人消费贷不良资产包。 这一举动，标志着这家以撮合交易闻名的投行，正式挺进不良资产处置这一"重资产"战场。 21世纪经济报道记者了解到， 除早年已布局的互联网平台外，近年来包括大型产业资本在内 的诸多巨头也纷纷入场不良资产处置。 其中采用的一种常见的模式是： 由实力雄厚的投资方 作为资金提供方，委托具备专业尽调、竞价和处置能力的地方资产管理公司作为运营方在前端 市场筛选、竞逐资产包，并按约定共享收益。 而此次华兴资本的入局，成为市场变化中的又一个关键注脚。 一笔"划算"的买卖：3亿换7 4亿债权 此次华兴资本收购的 两笔不良资产包均来自奇富科技旗下、本息总额高达74.29亿元，而总对 价仅为3.08亿元，相当于以约4.15%的平均折扣率购入。 这两个资产包均属于典型的个人消费 信贷不良债权，具备此类资产普遍的特征：逾期时间长、无抵押、回收高度依赖催收与司法处 置。 具体来看，第一个资产包未 ...

这一举动，标志着这家以撮合交易闻名的投行，正式挺进不良资产处置这一"重资产"战场。 21世纪经济报道 记者郭聪聪 2025年12月31日，曾以主导滴滴与快的合并、美团与大众点评合并等重磅交易闻名的"并购之王"华兴资本，宣布将通 过全资子公司以3.08亿元的代价，收购奇富科技旗下本息总额高达74.29亿元的个人消费贷不良资产包。 一笔"划算"的买卖：3亿换74亿债权 此次华兴资本收购的两笔不良资产包均来自奇富科技旗下、本息总额高达74.29亿元，而总对价仅为3.08亿元，相当于 以约4.15%的平均折扣率购入。这两个资产包均属于典型的个人消费信贷不良债权，具备此类资产普遍的特征：逾期 时间长、无抵押、回收高度依赖催收与司法处置。 21世纪经济报道记者了解到，除早年已布局的互联网平台外，近年来包括大型产业资本在内的诸多巨头也纷纷入场不 良资产处置。其中采用的一种常见的模式是：由实力雄厚的投资方作为资金提供方，委托具备专业尽调、竞价和处置 能力的地方资产管理公司作为运营方在前端市场筛选、竞逐资产包，并按约定共享收益。 而此次华兴资本的入局，成为市场变化中的又一个关键注脚。 具体来看，第一个资产包未偿本金66.77亿 ...

Core Insights - The Beijing Financial Technology Industry Alliance announced the results of the "Top Five Group Standards" for 2025, recognizing the contributions of Qifu Technology in the standardization of financial large models [1][5] - Qifu Technology's participation in the development of two important group standards, namely "Technical Requirements for Financial Applications of Large Language Models" and "Evaluation Specifications for Financial Applications of Large Language Models," has been acknowledged as a significant achievement [1][3] Group 1: Standardization Efforts - Qifu Technology, as a core participating unit, has leveraged nine years of experience in financial technology to provide essential support across dimensions such as technology implementation, risk control, and business adaptation [3] - The standards were led by Industrial and Commercial Bank of China and involved collaboration with major industry players like China Mobile, Huawei, Tencent, and Alibaba, addressing the lack of unified norms in the application of large models in the financial sector [3][4] - The standards define core requirements including application framework, task capabilities, and security trustworthiness, establishing a comprehensive evaluation system for financial institutions' model construction, assessment, and risk management [3][4] Group 2: Company Initiatives - Qifu Technology is one of the early entrants in the financial large model sector, consistently advancing both "technology research" and "standard development" in tandem [4] - In 2023, the company formed a specialized large model team and collaborated with the China Academy of Information and Communications Technology to compile the first financial industry large model standard [4] - The company aims to launch a super intelligent agent in 2025 that enhances core lending operations by integrating modules for credit decision-making, credit assessment, and compliance assistance, thereby providing a practical example of standardized technology implementation [4][5] Group 3: Future Directions - The recognition as one of the "Top Five Group Standards" is seen as a positive affirmation of Qifu Technology's efforts in promoting industry standardization [5] - The company plans to continue collaborating with academic and research partners to deepen technological research and practical transformation, actively participating in the construction of industry standard ecosystems [5]

Core Viewpoint - The article discusses the challenges and opportunities of integrating AI models into the financial credit assessment process, emphasizing the need for a standardized evaluation framework to measure AI performance in real-world scenarios [2][4][10]. Group 1: Challenges in AI Credit Assessment - AI models struggle with real-world data complexities, such as poor image quality and non-standardized documents, which can hinder their effectiveness in credit assessments [2][3]. - The financial industry lacks a unified benchmark to evaluate AI models, leading to anxiety among institutions when selecting appropriate tools [4][5]. - There is a misalignment between the capabilities of existing AI models and the specific requirements of credit assessment tasks, which often focus on nuanced document verification [6][8][10]. Group 2: Development of Evaluation Standards - The need for a tailored evaluation standard for AI in credit assessment is highlighted, which should be both industry-specific and technically robust [11][12]. - A collaborative effort between financial institutions and academic partners aims to create a comprehensive evaluation framework, FCMBench-V1.0, to address the unique challenges of credit assessment [16][18]. - The evaluation framework incorporates real-world data simulations to ensure that AI models are tested under conditions that closely resemble actual operational environments [18][20]. Group 3: Performance of AI Models - The FCMBench evaluation framework assesses AI models based on perception, reasoning, and robustness, ensuring they can handle complex credit assessment tasks [20][25]. - The Qfin-VL-Instruct model developed by Qifu Technology achieved the highest scores in the evaluation, demonstrating the effectiveness of specialized models over general-purpose ones in financial contexts [31][32]. - The Qfin model not only excels in accuracy but also offers improved speed and efficiency, making it suitable for real-time credit assessment scenarios [33]. Group 4: Future Outlook - The article emphasizes the importance of practical applications of AI in finance, suggesting that successful models must be grounded in real-world data and scenarios [36][37]. - Qifu Technology's initiative to open-source the FCMBench dataset and evaluation methods aims to bridge the gap between academia and industry, providing valuable resources for developing compliant and high-quality credit assessment tools [35][38].

Core Insights - The 2025 KPMG China FinTech Dual 50 Report marks the 10th anniversary of the selection, showcasing the industry's development during the critical period of the "14th Five-Year Plan" [1] - FinTech is transitioning from "digitalization" to "intelligentization," becoming a vital engine for serving the real economy, with "pragmatism" and "deepening" as the main themes of industry development [1][2] - The report highlights a significant concentration of companies in major urban areas, with Beijing, Shanghai, and Shenzhen leading the first tier, and the Yangtze River Delta, Guangdong-Hong Kong-Macau, and Beijing-Tianjin-Hebei regions accounting for 88% of the total [1][2] Company Composition - 90% of the listed companies have been established for over five years, while the proportion of companies founded within the last three years has increased to 6%, indicating a collaborative development between established and emerging players [1] - Over 80% of the listed companies have more than 40% of their workforce in technology roles, emphasizing the importance of core technical talent as a support for industry innovation [1][2] Technology Application - Artificial intelligence continues to lead, with 92% of the listed companies utilizing technological elements, collaborating deeply with big data and blockchain technologies, and penetrating core scenarios such as investment research and risk control [2] - The application of large models and intelligent agents is moving beyond conceptual phases, with a "collaborative model" reducing costs and improving response times, while multi-agent collaboration significantly enhances the accuracy of complex task handling [2] Industry Trends - FinTech services are penetrating the entire lifecycle of technology companies, utilizing intelligent credit assessments to meet diverse financing needs at different stages [2] - The industry is entering a 2.0 era of going global, forming a "dual market" model that promotes inclusive financial services in emerging markets while building competitive advantages through technology exports in mature markets [2] Capital Market Insights - 63% of the listed companies have IPO plans, with Hong Kong and domestic markets being the primary destinations for listings, and some companies adopting multi-location listing strategies [2] - As technological innovation deepens and regulatory frameworks improve, FinTech is expected to continue advancing in core technological breakthroughs, application scenario expansions, and enhancements in self-controllable capabilities, injecting lasting momentum into high-quality industry development [2]

Core Insights - Qifu Technology, in collaboration with Fudan University and South China University of Technology, has launched the first multimodal evaluation benchmark for credit scenarios, named FCMBench-V1.0, aimed at enhancing AI applications in credit assessment [1][2] - The benchmark is designed to address real credit business scenarios, focusing on multi-modal evaluation tasks that reflect practical business needs, thereby promoting academic research and application in credit AI [1] - Unlike traditional evaluations that focus on single recognition or understanding capabilities, this benchmark assesses model capabilities relevant to key processes in micro-enterprise credit granting, such as multi-document recognition and risk clue discovery [1] Summary by Sections - **Benchmark Development**: The FCMBench-V1.0 provides a standardized evaluation platform to foster collaboration between academia and industry, enabling fair comparisons of AI model capabilities in the credit field [1] - **Data and Framework**: The benchmark includes a highly consistent evaluation framework with 18 core credit document types, comprising 4,043 compliant images and 8,446 test samples, covering the entire credit review process [2] - **Industry Impact**: This initiative aims to break down data and knowledge barriers within the industry, facilitating a shift from "single-point optimization" to "collaborative innovation" between academia and financial technology companies [1]