Google Research Introduces Gemini-SQL2 with 80.04% BIRD Benchmark Accuracy

Google

Jun 15, 2026

Google Research introduced Gemini-SQL2, a text-to-SQL capability powered by Gemini 3.1 Pro. It achieved 80.04% execution accuracy on the BIRD benchmark, setting a new state-of-the-art for translating natural language into SQL queries that execute successfully. This capability addresses the challenge of generating accurate, runnable code from complex business contexts and natural language inputs.

BIRD Text-to-SQL Leaderboard (Single Model); Gemini-SQL2 Google 80.04%; Databricks RLVR 32B Databricks; Q-SQL AWS; SiriusAI-Text2SQL-32B-v2 Tencent; Gemini-SQL Google; GPT-5.5-xhigh OpenAI; Claude Opus 4.6 Anthropic; SQLWeaver-32B Alibaba; Arctic-Text2SQL-R1-32B Snowflake — Gemini-SQL2 achieves top performance on the BIRD Text-to-SQL benchmark with an execution accuracy of 80.04 percent.

View the full update on x.com

Google Research

@GoogleResearch2d ago

🚀 Introducing Gemini-SQL2, our breakthrough text-to-SQL capability powered by Gemini 3.1 Pro! We've achieved state-of-the-art results on the highly competitive BIRD benchmark, translating natural language into execution-ready SQL queries. 🧵👇 https://t.co/HfO2ZW2pih

6146.7k

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Keep reading

Gemini 3.1 Pro Tops ARC-AGI-2 Benchmark with 77.1% Score

Google released Gemini 3.1 Pro, scoring 77.1% on ARC-AGI-2 - more than double Gemini 3 Pro's score and ahead of Claude Opus 4.6. It's rolling out to the Gemini app, NotebookLM, and developer APIs including Gemini CLI and Antigravity.

Google Research Benchmarks Gemini's 3D Object Generation Through Code

GoogleJun 7

Google Research Benchmarks Gemini's 3D Object Generation Through Code

Google Research introduced 3DCodeBench, a new benchmark evaluating AI models' ability to generate 3D objects using code. This benchmark, presented at CVPR2026, demonstrates how agentic AI can autonomously create complex 3D assets, highlighting the role of iterative refinement in improving model performance.

Google Gemma 4 Claims Top Rankings on Japanese Swallow Leaderboard v2

Google GemmaMay 12

Google Gemma 4 Claims Top Rankings on Japanese Swallow Leaderboard v2

Google confirmed that its Gemma 4 open-weight model achieved high-ranking results on the Swallow Leaderboard v2, a rigorous Japanese language benchmark. This validation establishes the model as a leading choice for developers building regional applications that require frontier-level reasoning in Japanese.

Google Gemini 3.1 Flash Live Claims Top Spot for Production Voice Agents

Google AI StudioApr 13

Google Gemini 3.1 Flash Live Claims Top Spot for Production Voice Agents

Google's Gemini 3.1 Flash Live model reached the #1 position on the Tau Voice Bench leaderboard for real-time voice agents. The update delivers significantly lower latency and higher precision, signaling that multimodal voice AI is now reliable enough for production-grade applications.