via MarkTechPost
Google Releases Gemini-SQL2: Gemini 3.1 Pro Text-to-SQL Scores 80.04% on BIRD Single-Model Leaderboard
Google Research has announced the launch of Gemini-SQL2, a breakthrough text-to-SQL system powered by Gemini 3.1 Pro. According to an official post on X, the system achieved 80.04% execution accuracy on the BIRD Text-to-SQL Leaderboard (Single Model category), surpassing its predecessor Gemini-SQL and claiming the top spot.
Execution accuracy measures whether the generated SQL runs successfully and returns correct results, rather than simply checking syntactic validity. This metric reflects real-world robustness for converting natural language questions into precise database queries.
By early 2026, the BIRD benchmark has become the de facto standard for evaluating text-to-SQL systems, emphasizing both correctness and efficiency. Gemini-SQL2's performance marks a significant improvement over previous state-of-the-art models, narrowing the gap between AI-generated and human-written SQL.
Google's approach combines advanced reasoning capabilities in Gemini 3.1 Pro with optimized prompt engineering and schema linking strategies. The system handles complex queries involving multi-table joins, nested subqueries, and domain-specific functions, addressing longstanding challenges in the field.
This release signals growing enterprise readiness for conversational database interfaces. As organizations continue to accumulate data, solutions like Gemini-SQL2 could democratize data access, allowing non-technical users to perform sophisticated analytics without SQL expertise.
The BIRD leaderboard update places Gemini-SQL2 ahead of other single-model approaches, though ensemble methods and specialized agents remain competitive. Google has not yet announced public availability or API access for Gemini-SQL2.