Google Releases Gemini-SQL2; Gemini 3.1 Pro Scores 80.04% on BIRD

Google Research announced Gemini-SQL2 on June 12, 2026: a text-to-SQL capability powered by Gemini 3.1 Pro that posted 80.04% execution accuracy on the BIRD single-model leaderboard. The announcement highlights that execution-accuracy number as the core performance metric and frames Gemini-SQL2 as a schema-grounded text-to-SQL approach intended for practical database query generation.

The reported 80.04% figure is the BIRD benchmark’s execution-accuracy result for Gemini 3.1 Pro on the single-model leaderboard; the source material explains what that metric measures and how leaderboard placement is assessed. At the same time, Google has not yet published full implementation or evaluation details needed to reproduce the benchmark or immediately integrate the capability into production systems. The release also discusses use cases and a schema-grounded implementation pattern, which suggests recommended integration approaches even in the absence of full technical disclosure.

For engineers and product teams, the immediate takeaway is that Gemini-SQL2 establishes a public performance point for Gemini 3.1 Pro on a recognized text-to-SQL benchmark, but adoption decisions will depend on further details. Expect to watch for follow-up releases from Google that provide evaluation configurations, prompts or templates, APIs or SDKs, and multi-model comparisons that enable reproducible testing and practical deployment planning.