Ant Group Launches Financial Reasoning AI Model

At the World Artificial Intelligence Conference, Ant Digital unveiled its latest financial reasoning large model, Agentar-Fin-R1. Built on Alibaba's Tongyi Qianwen Qwen3, this model is designed to address the financial industry's need for professionalism, reasoning capabilities, and secure compliance.

Performance and Capabilities

Agentar-Fin-R1 has outperformed open-source general large models like Deepseek-R1 on authoritative benchmarks such as FinEval1.0 and FinanceIQ. This achievement marks a significant milestone in AI applications for the financial sector.

Tailored Solutions for Financial Institutions

To cater to diverse scenarios, Ant Digital offers two versions of the model:

• 32B and 8B parameter versions for reasoning tasks.
• 14B and 72B parameter versions for non-reasoning applications.

Zhao Wenbiao, CEO of Ant Digital, emphasized the "knowledge gap" between general AI models and industrial needs. He stated that financial large models are essential for the deep integration of AI in finance.

Data and Expertise

The model leverages:

• A comprehensive financial task classification system (6 categories, 66 subcategories).
• Trillions of financial professional data corpus.
• Expert-annotated financial long reasoning chains (CoT).

These resources enable Agentar-Fin-R1 to demonstrate an "innate understanding of finance."

Real-World Impact

Ant Digital has already deployed the model in a Shanghai bank's "AI mobile banking" application. This innovation:

• Provides natural dialogue-based financial services.
• Improves satisfaction among elderly customers.
• Boosts monthly active users by 25%.

Key Points

• Agentar-Fin-R1 is built on Alibaba's Tongyi Qianwen Qwen3.
• It outperforms competitors on financial benchmarks like FinEval1.0.
• Offers tailored versions for different institutional needs.
• Leverages extensive financial data and expert annotations.
• Already enhancing user experience in real-world banking applications.