Ant Group Digital Technologies Open-Sources Agentar-SQL After Topping Global SQL Rankings for Over Two Months
Want to read in a language you're more familiar with?
Ant Group Digital Technologies has open-sourced its top-ranked Agentar-SQL technology after dominating global NL2SQL benchmarks for more than two months.
December 14, 2025 — At the Second CCF China Data Conference, Ant Group Digital Technologies (Ant Digital) announced the open-sourcing of its Agentar-SQL series, a core technology powering data intelligence agents.
The open-source release includes full research papers, source code, model weights, and user documentation for its real-time text-to-SQL (NL2SQL) framework. Ant Digital also revealed plans to gradually open-source end-to-end technical frameworks, covering database understanding and data mining capabilities.
In September 2025, Ant Digital’s Agentar-Scale-SQL achieved first place on BIRD-Bench, a globally recognized NL2SQL evaluation benchmark. It recorded 81.67% execution accuracy and 77% execution efficiency, outperforming major international players such as Google and Amazon, and held the top position for more than two consecutive months.
The technology addresses four major challenges in real-world NL2SQL deployment. During pilot operations at a leading city commercial bank in China, the tool achieved an average query accuracy exceeding 92%, delivering more than a threefold improvement over traditional solutions.
Ant Digital has long focused on AI applications in the financial sector, serving 100% of China’s state-owned and joint-stock banks and over 60% of regional commercial banks. It has also partnered with institutions such as Bank of Ningbo and Bank of Shanghai to co-develop intelligent financial systems.
More recently, Ant Digital’s capabilities have expanded beyond finance into public services and energy. In collaboration with Nanjing Public Transport, it launched an AI-powered transit agent called “Little Blue Whale,” which has helped introduce over 30 bus routes based on AI-driven recommendations. In the energy sector, its EnergyTS foundation model has reduced investment evaluation timelines from 2–3 days to just over 10 minutes.
Ant Digital also adopts a “pay-for-performance” commercial model and continues to upgrade its “Stellar Program” ecosystem initiative. To date, the company has served more than 100 overseas financial institutions, highlighting the growing global influence of China’s industrial AI technologies.
