Fin-R1 financial model: 7B parameters defeat industry giants, open source helps financial technology

The Fin-R1 model jointly developed by Professor Zhang Liwen's team (SUFE-AIFLM-Lab) and Caiyuexingchen, School of Statistics and Data Science, Shanghai University of Finance and Economics, was officially open source. This financial-specific big model based on Qwen2.5-7B performed well in several financial benchmarks through reinforcement learning training.

With its 7B parameter scale, Fin-R1 has performed well in tasks such as Financial Table Inference (FinQA) and Conversational Financial QA (ConvFinQA), surpassing many larger-scale competitors. This model is designed for core financial business scenarios. It can write financial code, build pricing models, conduct quantitative analysis, generate professional reports, provide security compliance analysis, and realize intelligent risk control and ESG sustainability analysis.

Fin-R1 is built on Qwen2.5-7B-Instruct, and uses the DeepSeek-R1 framework for data distillation and quality screening. Combined with supervised fine-tuning (SFT) and reinforcement learning (RL) training methods, it supports the Chinese and English environment and provides strong support for the digital transformation of the financial industry.