On November 8, YIXIN (02858) unveiled the auto finance industry's first Agentic large model—XinMM-AM1—at the 2025 World Internet Conference Wuzhen Summit. This model significantly enhances customer acquisition, risk control, and operational efficiency while improving user experience, positioning itself as the "core engine" driving auto finance into the next decade.

Jia Zhifeng, Chief Technology Officer of YIXIN, officially introduced the Agentic large model at the Wuzhen event. The XinMM-AM1 boasts approximately 30 billion parameters, with a response latency below 200ms, supporting real-time voice agent interactions and achieving a single-card throughput of 370 tokens/s, enabling cost-effective large-scale deployment.

Trained on over 15T tokens of data—primarily sourced from YIXIN’s diverse and representative business scenarios—the model serves as the "core brain" and orchestrator for operations. It integrates key capabilities such as omnichannel interaction, multimodal perception, global collaboration, and full-scale security compliance, empowering intelligent and automated decision-making across the entire auto finance chain—from customer acquisition and application processing to smart risk control, fund routing, AI customer service, and asset management.

By enhancing risk control and operational quality, XinMM-AM1 significantly improves financing approval rates while addressing industry challenges like lengthy cycles, complex interactions, and multifaceted decision-making.

As an AI-driven fintech platform, YIXIN has invested over RMB 2 billion in R&D. In 2024, it became China’s first auto finance company to secure regulatory approval for a generative AI large model, achieving full-scenario AI deployment. In March this year, YIXIN open-sourced the industry’s first high-performance inference model, contributing to technological advancements in the sector.

With the launch of its Agentic large model, YIXIN aims to accelerate AI adoption in the industry and further advance the development of intelligent auto finance ecosystems.