Google Research has unveiled Gemini-SQL2, a new text-to-SQL system built on Gemini 3.1 Pro, which translates natural language into executable database queries. On the BIRD benchmark, the model achieves an execution accuracy of 80.04 percent, placing it ahead of OpenAI’s GPT-5.5-xhigh at 72.8 percent and Anthropic’s Claude Opus 4.6 at 70.9 percent. Other models from Databricks, AWS, Tencent, and Alibaba trail significantly behind these figures. The system is designed to handle the inherent difficulty of converting unstructured language into precise SQL, particularly when data is layered or requires complex business logic. Google states that the generated queries are both syntactically correct and capable of successful execution.
This advancement matters because reliable text-to-SQL capabilities are essential for improving natural language features across Google’s broader data services. Accurate translation reduces the barrier for non-technical users to interact with databases, potentially automating routine data retrieval and analysis tasks. However, the research team has not announced a public release of the model, nor has a technical paper been published yet. This lack of transparency suggests the technology may remain an internal tool for now, limiting immediate external verification or adoption by third parties. Until further details emerge, the practical impact on the wider industry remains theoretical rather than operational.
- Gemini-SQL2 leads current benchmarks with 80.04 percent execution accuracy on the BIRD test set.
- Competing models from OpenAI, Anthropic, and major cloud providers lag by several percentage points.
- No public release or academic paper is available, restricting independent evaluation of the technology.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
