Google Launches Gemini-SQL2, Hits 80% Accuracy in Text-to-SQL

Google Research has launched Gemini-SQL2, a model built on Gemini 3.1 Pro for Text-to-SQL tasks. It turns natural language questions into SQL queries, so users don't need to write complex database commands. In the BIRD benchmark, Gemini-SQL2 scored 80.04% accuracy on single model track, beating its earlier version. BIRD includes 95 databases, 37 fields, and over 12,000 question-SQL pairs, with messy data and external knowledge needs, making it more like real business use. Business people can now use plain language to check revenue or customer churn. Google hasn't shared the model's ID, API, or which products will use it first.