Ant Group's Ant Financial Digital Technology releases a large financial reasoning model, accelerating the "long - distance run" of financial intelligent agents | Frontline News

Author | Wang Fangyu

Editor | Su Jianxun

As the cornerstone of the new generation of artificial intelligence, large AI models provide a powerful technological foundation for the intelligent transformation of the financial industry, lowering the threshold for development and application, and can be regarded as the "new infrastructure" for financial intelligence.

On July 28th, at the World Artificial Intelligence Conference (WAIC 2025), Ant Digital Technology officially launched the large financial reasoning model Agentar-Fin-R1.

It is reported that this model is specially designed for financial industry applications and has achieved the highest scores in the three major mainstream financial benchmark tests, surpassing open-source general large models such as Deepseek and other financial large models, and showing outstanding performance in financial professionalism, reasoning ability, and security and compliance capabilities.

In addition, Ant Digital Technology also provides non-reasoning versions of 14B and 72B parameter large financial models, as well as a large model with a MOE architecture based on the Bailing foundation large model to meet the needs of financial institutions for deployment in diverse scenarios.

Currently, large AI models in financial scenarios still face challenges such as the hallucination problem of large models, output stability issues, and process interpretability problems. Therefore, specialized large financial reasoning models have become an inevitable requirement.

Based on the pain points in the industry, Wang Wei, the CTO of Ant Digital Technology, told 36Kr that Ant Digital Technology's model product strategy is to evolve from horizontal general large models to professional/specialized models. "We can draw on the knowledge of customers, scenarios, and solutions accumulated in past products more deeply, and then achieve a transformation to vertical specialization."

Specifically, at the data level, Ant Digital Technology has built a comprehensive and professional financial task classification system, including 6 major categories and 66 sub-categories of scenarios, covering the entire financial spectrum such as banking, securities, insurance, funds, and trusts. Based on hundreds of billions of financial professional data corpora, through trusted data synthesis technology and a construction mechanism that combines expert-labeled financial long thinking chains (CoT), the model's ability to handle complex tasks has been significantly improved.

At the training level, Ant Digital Technology has adopted an innovative weighted training algorithm to improve the learning efficiency and performance of large models for complex financial tasks. In subsequent business applications, it can significantly reduce the data requirements and computing power consumption for secondary fine-tuning, lowering the threshold and cost for the implementation of large models.

It is worth mentioning that Ant Digital Technology not only conducts research and development on industry models but also deploys a full-stack solution from large financial industry models, AI platforms to upper-layer applications.

For example, if a large model is the "brain", then an intelligent agent is the "body" and "actor" that transforms the cognitive ability of this "brain" into specific financial business execution capabilities. The large financial reasoning model launched by Ant Digital Technology helps to further accelerate the implementation and application of financial intelligent agents.

It is reported that currently, Ant Digital Technology has jointly launched over a hundred financial intelligent agent solutions with partners in the financial industry, covering four major fields including banking, securities, insurance, and general finance. Financial institutions can use these solutions "plug and play", increasing the work efficiency of front-line employees by over 80%.

For large AI models to truly become the key engine driving business growth, it is not only about technological breakthroughs but also about understanding and practicing in financial scenarios. Ant has long-term practical experience in the latter aspect.

Zhang Peng, the person in charge of Ant Digital Technology's AI technology, said that Ant Digital Technology stands at the intersection of large AI models and finance. Ant Group has many self-practiced business scenarios. AI intelligent agents have been tested in Ant's own practice, and at the same time, Digital Technology can also interact with different types of financial institutions, thus accumulating rich cross-experience.

Not only Ant Digital Technology, but at the World Artificial Intelligence Conference (WAIC 2025), technology manufacturers' intelligent agent application solutions in the financial field have blossomed everywhere. From the perspective of application scenarios, financial intelligent agents have also moved from "single-point attempts" as customer assistance tools to core business scenarios such as credit decision-making, gradually moving towards large-scale application.

"Currently, it is an era of a hundred flowers blooming for AI intelligent agents, and the development of AI intelligent agents will be a long-distance race. The path that Ant Digital Technology wants to take is to keep running in the vertical field, especially in the financial field, to give full play to Ant's advantages to a greater extent." Wang Wei told 36Kr.