How Ant Group’s Agentar‑Fin‑R1 Redefines Financial AI with Expert‑Level Reasoning

On July 28 at the World Artificial Intelligence Conference, Ant Group’s Ant Financial Science announced the financial reasoning large model Agentar‑Fin‑R1, designed as a reliable, controllable, and optimizable AI core for finance.

While the financial industry is rapidly adopting large models, real‑world scenarios demand deep domain expertise, complex logical reasoning, and strict security and compliance, which generic models often cannot satisfy.

Agentar‑Fin‑R1 is built specifically for finance and offers three major advantages:

Domain expertise out‑of‑the‑box : Leveraging a comprehensive taxonomy of 6 major and 66 sub‑categories covering banking, securities, insurance, funds, trust, and more, the model is trained on billions of finance‑specific data points and expert‑annotated chain‑of‑thought (CoT) reasoning, making it a “factory‑ready expert.”

Efficient training algorithm : An innovative weighted training method improves learning efficiency for complex financial tasks, dramatically reducing the data and compute needed for downstream fine‑tuning, thus lowering deployment cost.

Continuous self‑evolution : The model can ingest the latest financial regulations and market dynamics, and is regularly evaluated with dedicated tools to optimize performance in real‑world applications.

To assess its capabilities, Ant Financial Science partnered with Industrial and Commercial Bank of China, Ningbo Bank, Beijing Frontier Financial Regulation Technology Research Institute, and Shanghai AI Industry Association to launch the Finova financial model benchmark, testing reasoning, intelligence, and compliance.

The open‑source Finova benchmark is now publicly available, encouraging industry‑wide improvement of financial LLMs. Evaluation results show that Agentar‑Fin‑R1 surpasses open‑source counterparts such as Deepseek‑R1 on FinEval1.0, FinanceIQ, and Finova, demonstrating superior financial expertise, reasoning, and compliance while maintaining strong general‑purpose abilities.

In addition, Ant Group has co‑founded the “Financial Intelligent Agent Application Co‑Creation Alliance” with more than ten technology partners to promote standards, ecosystem collaboration, and widespread adoption of financial AI.

Looking ahead, Ant Group plans to deepen partnerships to transition large models in finance from merely “usable” to truly “useful,” accelerating large‑scale value creation across the sector.