CNFinBench Releases Comprehensive Scores for Large Language Models: Ping An Financial Large Model Ranks First

(Source: Taianet)

Repost from: Taianet

Recently, China Ping An’s financial large model PingAnGPT-Qwen3-32B ranked first overall on the industry-authoritative large model evaluation platform CNFinBench’s public leaderboard for large language models. The models participating in this evaluation are highly competitive and diverse, including open-source large models such as DeepSeek-R1 (671B), Qwen3-235B-A22B, Kimi-K2 (1000B), as well as leading closed-source commercial models like GPT-4o, Claude-Sonnet 4, Doubao-1.5-pro, representing the industry’s highest standards. Achieving the top spot in this evaluation also marks that Ping An’s technological capabilities in the field of financial large models have reached an advanced industry level.

CNFinBench is a comprehensive evaluation benchmark for Chinese financial large models developed by the Shanghai Artificial Intelligence National Laboratory team in collaboration with authoritative financial institutions. It is currently the most authoritative and industry-recognized platform for assessing financial large model capabilities in China. In this evaluation, PingAnGPT-Qwen3-32B demonstrated outstanding performance in key areas such as financial fact reasoning and calculation, financial professional knowledge Q&A, financial compliance and risk control, showcasing precise financial numerical computation, rigorous logical reasoning, and a deep understanding of the entire financial knowledge system. It is especially valuable in scenarios like financial research analysis and risk measurement, offering significant application value and safety advantages.

The evaluators stated that topping CNFinBench reflects Ping An’s strong technical strength in the field of financial large models. Notably, the model from China Ping An uses only 32B parameters, less than 5% of DeepSeek-R1’s (671B) and 3.2% of Kimi-K2’s (1000B), yet surpasses them in overall performance, achieving a major breakthrough in balancing model efficiency and performance. Thanks to its “lightweight and efficient” design, this model supports 97 business scenarios including property and auto insurance claims reporting, HR intelligent query, customer service real-time quality inspection, intelligent dialogue, and group financial expense review, laying a solid foundation for future private deployment and scenario-based applications.

It is reported that the CNFinBench evaluation system covers five core dimensions: financial professional knowledge Q&A, financial business understanding and analysis, financial fact reasoning and calculation, financial compliance and risk control, and financial endogenous and application security. It comprehensively tests the model’s professional depth, business understanding, calculation accuracy, compliance awareness, and safety in the financial vertical.

In the first three quarters of 2025, Ping An’s database accumulated 30 trillion bytes of data, covering nearly 250 million individual customers. By training large models on massive data, the company has accumulated over 3.2 trillion high-quality text corpora, 310,000 hours of annotated speech data, and over 7.5 billion images. Relying on vast data resources and technological services, Ping An continues to expand the depth and breadth of scenario applications, empowering financial services with better experiences, risk control, cost reduction, and sales promotion. In February 2026, the National Intellectual Property Administration released the “Patent Analysis White Paper for the Financial Technology Industry 2025” and the “Patent Analysis White Paper for the Healthcare Industry 2025,” with China Ping An again ranking first in patents in both the financial technology and healthcare sectors, leading in multiple core technologies and key application scenarios.

China Ping An states that by building leading AI capabilities, promoting model optimization and iteration, and deepening scenario-based application deployment, the company continuously transforms technological advantages into service value. Through high-quality digital financial services, it aims to meet the people’s aspirations for a better life, fulfilling its service promise of “peace of mind, time-saving, and cost-saving,” and contributing to the construction of a strong financial nation.