Google Releases Gemini-SQL2: Gemini 3.1 Pro Text-to-SQL Scores 80.04% on BIRD Single-Model Leaderboard

Editor's take

Google Research has introduced Gemini-SQL2, a text-to-SQL system leveraging Gemini 3.1 Pro, achieving 80.04% execution accuracy on the BIRD leaderboard.

This development is significant as it marks a continued push for large language models to reliably translate natural language queries into structured database commands. The BIRD leaderboard, which evaluates single-model performance, is a crucial benchmark for assessing the practical utility of LLMs in data analysis and retrieval tasks, directly impacting businesses reliant on efficient data access.

Future attention should focus on Gemini-SQL2's performance against ensemble models on BIRD, and its ability to handle increasingly complex SQL dialects and schema variations. A sustained improvement in accuracy, particularly in edge cases and for specialized databases, will be key to its broad adoption.