On September 26th, the official website of the globally recognized evaluation benchmark, BIRD-Bench, revealed that Ant Group Digital Technologies' data analysis agent, Agentar-SQL, clinched the first position worldwide in both execution accuracy (scoring 81.67 points) and execution efficiency (with 77 points). This remarkable feat places it ahead of industry giants such as AT&T, Google Cloud, Tencent Cloud, and Alibaba Cloud. The evaluation demands that AI models accurately transform natural language queries into SQL statements that can be reliably executed in real-world, complex databases. The dataset encompasses 37 industry-specific scenarios, with a total size of 33GB. Agentar-SQL is constructed on Ant Group Digital Technologies' proprietary SQL large model and utilizes the GSPO reinforcement learning training approach. It incorporates a multi-round reflection and correction mechanism, along with a two-stage generation method, significantly boosting the precision and efficiency of SQL generation.