Ant Launches Finance-Focused Large Model as AI Gains Ground in Banking
At the recent World Artificial Intelligence Conference (WAIC) on July 28, Ant Digital Technologies introduced Agentar-Fin-R1, a large AI model built specifically for financial reasoning. This marks a clear shift from experimental AI applications to broader deployment within the financial sector.
Agentar-Fin-R1 scored highest on three major industry benchmarks, outperforming not only general-purpose open-source models but also specialized financial models like DeepSeek. The model’s strengths lie in its deep domain expertise, advanced reasoning capabilities, and built-in compliance safeguards—critical features for financial institutions.
Specialized Models to Meet Diverse Needs
In addition to Agentar-Fin-R1, Ant released non-reasoning financial models with 14 billion and 72 billion parameters, plus a mixture-of-experts (MoE) model based on the Bailing foundation model. These options offer flexibility for different deployment scenarios.
Despite improvements, challenges remain with foundation models in finance, including hallucination, inconsistent outputs, and limited explainability. Agentar-Fin-R1’s design specifically addresses these issues, aiming for more reliable and transparent AI-driven decisions.
Built from Deep Financial Experience
Wang Wei, CTO of Ant Digital Technologies, emphasized the move from generic to specialized AI solutions. Ant leverages years of client collaborations and real-world financial scenarios to create models that deliver real value.
The company has developed a detailed task taxonomy covering six major categories and 66 subcategories across banking, securities, insurance, funds, and trusts. Training involved hundreds of billions of financial tokens combined with a chain-of-thought framework guided by domain experts. This approach improves performance on complex tasks while optimizing data efficiency and computing resources, reducing costs for deployment.
From AI Models to Practical Applications
Ant is building a comprehensive technology stack that spans foundation models, AI platforms, and finance-specific business applications. If the large model acts as the “brain,” the financial AI agent is the “body,” executing tasks in real-world settings.
Agentar-Fin-R1 is central to accelerating this transition. So far, Ant has partnered with financial institutions to deploy over 100 AI agent solutions across banking, securities, insurance, and other financial services. Results include more than 80% improvement in frontline staff efficiency.
Industry Momentum and Future Outlook
Ant's head of AI, Zhang Peng, highlighted the advantage of combining operational experience with AI development. Having tested these AI agents internally and with external partners, the company brings a thorough understanding of practical needs and constraints.
Other tech firms also showcased financial AI agents at WAIC 2025, signaling a wider industry move toward AI integration in core functions such as credit approval. Wang Wei noted, “We’re entering an era where AI agents are in full bloom. This will be a long-term effort with deep focus on verticals like finance, where Ant’s expertise stands out.”
For professionals in finance interested in practical AI applications, exploring specialized AI training can be valuable. Resources like Complete AI Training’s finance AI tools offer targeted courses to understand and implement these technologies effectively.
Your membership also unlocks:
