金融智能体平台Agentar

Search documents
都“百模大战”了,蚂蚁数科为何要发布金融推理大模型?
Tai Mei Ti A P P· 2025-08-05 07:13
金融行业又迎来了一个新的大模型。 7月底举行的2025世界人工智能大会(WAIC)论坛上,蚂蚁数科正式发布金融推理大模型Agentar-Fin- R1,提供32B和8B两个版本,为金融AI应用打造可靠可优化智能中枢。 该模型基于Qwen3研发,在FinEval1.0、FinanceIQ等权威金融大模型评测基准上超越同尺寸开源通用大 模型以及金融大模型,显示其更强的金融专业性、推理能力以及安全合规能力。 实际上,随着大模型从技术走向应用落地,金融行业的实际业务场景中,往往需要高度专业的金融知 识、复杂的业务逻辑推理能力以及严格的金融级安全合规等要求,然而,现有的基础大模型"幻觉率"依 然很高,在解决实际金融任务时仍然存在诸多挑战,因此,"百模大战"下,金融这一专业领域也需要更 懂金融的专业大模型。 蚂蚁数科CEO赵闻飙表示,通用大模型距离产业实际应用存在"知识鸿沟"。构建专业的金融大模型是推 进金融与AI深度融合的必然路径,未来,金融大模型的应用深度将成为金融机构竞争力的关键要素。 基础大模型无法适应金融专业场景 研究机构IDC于近日发布的一份报告显示,2024年,中国金融行业生成式 AI平台及应用解决方案市场 ...
蚂蚁数科CEO赵闻飙:破解产业难题 锻造专业金融大模型
Huan Qiu Wang· 2025-07-28 05:25
Core Viewpoint - Ant Group's subsidiary, Ant Financial Technology, has launched the financial reasoning model Agentar-Fin-R1, aimed at creating a reliable, controllable, and optimizable AI core engine for the financial sector [1][3]. Group 1: Model Development and Performance - Agentar-Fin-R1 is developed based on the domestic foundational model Qwen3 and has outperformed other open-source general models and specialized financial models in authoritative financial evaluations such as FinEval1.0 and FinanceIQ [1][3]. - The model demonstrates significant advantages in financial professionalism, complex reasoning capabilities, and safety compliance [1][3]. Group 2: Industry Challenges and Solutions - The financial industry's digital transformation is accelerating, but existing general models face challenges due to the need for specialized financial knowledge and stringent compliance standards [3]. - Ant Financial Technology emphasizes the necessity of building specialized financial models to bridge the knowledge gap between general models and industry needs, which will be a key indicator of financial institutions' core competitiveness [3]. Group 3: Model Features and Capabilities - Ant Financial Technology has established a comprehensive financial task classification system covering various sectors such as banking, securities, insurance, and trusts, divided into 6 major categories and 66 subcategories [4]. - The model utilizes a unique weighted training algorithm to enhance learning efficiency and performance, significantly reducing the data and computational power required for subsequent business fine-tuning [4]. - Agentar-Fin-R1 is designed for continuous iteration, capable of absorbing the latest financial policies and market dynamics for targeted optimization [4]. Group 4: Market Position and Applications - Ant Financial Technology is accelerating its enterprise-level model services, focusing on finance and new energy sectors, with its financial AI platform Agentar achieving the highest rating in evaluations by the China Academy of Information and Communications Technology [5]. - The company has successfully implemented AI solutions in various financial institutions, significantly improving user satisfaction and engagement, as evidenced by a 25% year-on-year increase in monthly active users for an AI mobile banking service [5].