Ant Digital Unveils Financial AI Model as China’s LLM Race Reaches Banking S...

AsianFin -- Ant Digital Technologies Co. has launched a new large language model tailored for financial reasoning, marking its latest bid to secure a leading role in China's increasingly specialized AI landscape.

The model, named Agentar-Fin-R1, was unveiled at the 2025 World Artificial Intelligence Conference in Shanghai last week. Developed on top of Qwen3, Ant's proprietary large model architecture, Agentar-Fin-R1 comes in 32 billion and 8 billion parameter versions, designed to serve as a “trustworthy, optimizable intelligent core” for financial industry applications.

The Hangzhou-based company said Agentar-Fin-R1 outperforms comparable open-source financial LLMs on benchmarks like FinEval 1.0 and FinanceIQ, addressing key pain points such as reasoning complexity, security compliance, and hallucination control — a persistent issue in applying general-purpose AI models to real-world finance tasks.

“There’s still a knowledge gap between foundation models and industry-level applications,” Zhao Wenbiao, CEO of Ant Digital, said. “Building verticalized financial models is inevitable for AI’s deep integration into banking, insurance, and asset management.”

As generative AI transitions from hype to deployment, financial institutions are finding that general-purpose models fail to meet the sector’s stringent demands for data privacy, regulatory compliance, and logic-based reasoning.

According to IDC, China’s market for generative AI platforms and application solutions in the financial sector is projected to reach 3.5 billion yuan ($480 million) by 2027, growing nearly fivefold from 2024. About 91% of this market is expected to come from on-premises deployments, given data security concerns.

Agentar-Fin-R1 is Ant’s answer to these demands. The model was built on a dataset encompassing six major financial sectors and 66 subcategories, spanning banking, securities, insurance, funds, and trusts. Ant leveraged hundreds of billions of financial data points and developed a chain-of-thought annotation system to enhance reasoning precision.

The company also introduced a weighted training algorithm to improve data efficiency, reducing the need for costly fine-tuning processes. This, Ant claims, will lower the barriers for financial institutions to integrate AI at scale.

Agentar-Fin-R1 topped FinEval 1.0 and FinanceIQ benchmarks, outperforming open-source rivals like DeepSeek and Xuanyuan. In addition, it led the Finova Large Model Financial Application Benchmark, co-developed with Industrial and Commercial Bank of China, Bank of Ningbo, and other institutions, which measures agent capabilities, complex reasoning, and security compliance.

Ant Digital is offering the model in multiple configurations — including a Mixture of Experts (MoE) variant for high-speed inference — to meet diverse deployment needs. The company also provides 14B and 72B non-reasoning versions optimized for other financial tasks.

“Reliable reasoning models are the engine that drive enterprise-level intelligent agents,” said Wang Wei, CTO of Ant Digital. “Without them, the entire AI ecosystem lacks traction.”

Ant Digital’s financial AI ambitions reflect a broader shift among Chinese tech firms towards domain-specific LLMs. With foundational models reaching diminishing returns in general capabilities, companies are now racing to capture vertical markets such as finance, healthcare, and manufacturing.

Since early 2025, Ant Digital has launched more than 100 intelligent agent solutions for financial scenarios in partnership with banks and insurers. These range from AI-powered mobile banking and smart customer service to automated risk control systems.

For instance, a Shanghai-based commercial bank partnered with Ant to roll out a conversational banking app that allows users to conduct transactions through natural language, boosting monthly active users by 25% year-on-year.

Ant also helped Dadi Insurance develop an AI operations platform, integrating data, computing, and application frameworks — a first for China’s insurance sector. The collaboration reduced deployment cycles by 80% and increased accuracy by 30%, according to company estimates.

Ant Digital now counts 100% of China’s state-owned and joint-stock banks and over 60% of regional banks among its enterprise clients.

AI+Finance: From “Usable” to “Easy to Use”

While generative AI still faces significant hurdles in professional sectors, Ant Digital believes the next phase of adoption hinges on how seamlessly models can be embedded into business workflows.

“The question for banks is no longer whether to adopt AI, but how to operationalize it effectively,” said Wang Zhiheng, President of Agricultural Bank of China, during a recent industry forum.

Wang Wei echoed this sentiment, noting that enterprise-level intelligent agents have entered a breakout phase in 2025. “This is a marathon with no finish line,” he said. “But the shift from ‘usable’ to ‘easy to use’ will be key in determining who leads the AI+finance revolution.”