直击WAIC丨蚂蚁数科发布金融推理大模型，评测效果超主流开源大模型

Core Insights - Ant Group's Ant Financial Technology has launched the Agentar-Fin-R1 financial reasoning model, designed to create a "reliable, controllable, and optimizable" intelligent core for financial AI applications [1] - The model, based on Qwen3, surpasses other open-source general models and financial models in key evaluations, demonstrating superior financial expertise, reasoning capabilities, and compliance [1] - Ant Group's CEO emphasized the necessity of specialized financial models to bridge the "knowledge gap" between general models and practical industry applications, indicating that the depth of application of financial models will be a key competitive factor for financial institutions [1] Data and Model Specifications - Agentar-Fin-R1 is available in two versions: 32 billion and 8 billion parameters, with additional models based on the MOE architecture for improved reasoning speed [2] - Non-reasoning versions of the model include 14 billion and 72 billion parameters to cater to diverse deployment needs of financial institutions [2] - Ant Group has developed a comprehensive financial task classification system with 6 major categories and 66 subcategories, covering various financial sectors such as banking, securities, insurance, funds, and trusts [1]