At the 2nd CCF China Data Conference, Ant Group Digital Technologies made a significant announcement: it is open-sourcing the complete collection of research papers and code for its pivotal data intelligence agent technology, Agentar SQL. The initial phase of this open-sourcing initiative encompasses a real-time Text-to-SQL framework. Further technology frameworks are slated to be open-sourced in a phased manner in 2026.

During a trial run at a prominent city commercial bank, several tools leveraging this technology attained an impressive average query accuracy rate of over 92%. This marks a substantial enhancement, with accuracy more than tripling compared to traditional methods.

On September 25th of this year, Ant Group Digital Technologies' data analysis agent, Agentar-Scale-SQL, claimed the top spot on the authoritative BIRD-SQL benchmark. It maintained its leading position on both lists for over two consecutive months.

Research institutions forecast continued growth in both the global and Chinese business intelligence markets. Nevertheless, there are notable disparities in the extent to which Chinese enterprises utilize related products. This presents challenges for the widespread adoption of the NL2SQL (Natural Language to SQL) industry.

Zhang Peng, the head of AI technology at Ant Group Digital Technologies, highlighted four major hurdles in implementing NL2SQL. To ensure industrial viability, a comprehensive capability stack must be developed. This stack should include offline expansion, human-machine interaction, and self-evolution capabilities.

Ant Group Digital Technologies intends to gradually open-source these capability modules. The first-phase online expansion framework has already been released and is garnering attention from developers.