Today, with the rapid advancement of artificial intelligence technology, enabling AI to accurately understand and process database query tasks has always been a key focus for the industry. On June 12, Google Research officially released a new model called Gemini-SQL2. This specialized model, built upon Gemini 3.1 Pro, demonstrates exceptional capabilities in handling "text-to-SQL" tasks, and it has successfully topped the authoritative evaluation rankings.

The so-called "text-to-SQL" task involves enabling computers to understand human everyday conversations and convert them into executable database query instructions. In scenarios such as enterprise applications, self-service data analysis, and SaaS platforms, this technology can greatly reduce the barriers for users interacting with complex databases. However, the complexity of database table structures, ambiguous field definitions, and complicated business logic have long been the main obstacles for AI in handling such tasks.

image.png

In this context, the advantages of Gemini-SQL2 are particularly prominent. According to the latest data from the industry benchmark testing platform BIRD, Gemini-SQL2 achieved an execution accuracy rate of 80.04% on the single-model track, successfully surpassing its previous version from Google. Notably, the BIRD evaluation set includes 95 databases from 37 professional fields, with more than 12,000 questions in total. It not only simulates real enterprise environments but also specifically includes test items with dirty data and those requiring external knowledge, making it highly valuable.

The application prospects of this model are very broad. In the future, enterprise employees will no longer need to master obscure code; they can simply ask questions in natural language, such as "What is the situation of regional sales loss in the last quarter?" The system will then automatically retrieve the data and generate accurate analytical reports.

Although the industry is full of expectations for the practical implementation of this model, as of now, Google has not yet announced the specific model identifier, API interface details, or detailed technical report for Gemini-SQL2, nor has it revealed which products will be the first to integrate this capability. How this top-tier AI model will subsequently change the workflow of data analysis remains something worth watching closely within the industry.