Editorial illustration for Gemini‑SQL2 leads BIRD benchmark with 80.04% execution accuracy
Gemini‑SQL2 leads BIRD benchmark with 80.04% execution...
Google's Gemini-SQL2 just hit 80.04% execution accuracy on the BIRD benchmark. That's a specific, hard number. For context, OpenAI's GPT-5.5-xhigh sits at 72.8%.
Claude Opus 4.6 is at 70.9%. The rest—from Databricks, AWS, Tencent, Alibaba—are far behind. This isn't some arcane academic score.
BIRD tests a brutal, real task: translating a messy human question into perfect SQL that runs against a messy, real-world database. The complexity is in the business logic. "Last quarter's top sellers" means filtering for Q3, joining tables, applying regional tax rules, excluding test accounts.
Miss one step and the query fails, or worse, returns plausible lies. An 80% success rate here is a threshold. It suggests moving past syntax mimicry into genuine comprehension of how data is structured and governed.
For Google, this is a direct upgrade path for BigQuery, Looker, its entire data stack. A user who can reliably ask "show me the projects over budget last month" in English doesn't need to learn SQL. They don't need to wait.
Google Research unveiled Gemini-SQL2, a new text-to-SQL system built on Gemini 3.1 Pro. It translates natural language into executable SQL database queries. On the BIRD benchmark, which measures how accurately these translations work, Gemini-SQL2 hits an execution accuracy of 80.04 percent, putting it in first place, according to Google.
That seven-point lead over OpenAI is the story. It isn't trivial. It points to a specific engineering advantage, likely in how Gemini-SQL2 is trained on the labyrinthine rules of actual enterprise data.
The others are now playing catch-up. This creates a tangible moat. Data work is tedious.
A tool that cuts through even part of that tedium with reliable, plain-English queries will get used. Google, with this result, has a clear technical claim to being that tool. The benchmark trophy is secondary.
The real win is making the case, with the hard number 80.04, that talking to your database is no longer science fiction. It's a feature.
Related Reading
Browse all Google news →
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
Related Topic Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI
From Archives NVIDIA tops AA‑AgentPerf benchmark, credits Vera Rubin platform
From Archives Perplexity routes deep‑research subtasks across 20+ models using Gemini agent
From Archives German Court Holds Google Liable for False AI-Generated Overviews
LLMs & Generative AI
From Archives Google's DiffusionGemma: open diffusion model for faster text generation
LLMs & Generative AI
Common Questions Answered
What is Gemini-SQL2's execution accuracy score on the BIRD benchmark?
Gemini-SQL2 achieved 80.04% execution accuracy on the BIRD benchmark, which represents the highest performance among all tested models. This score places it seven percentage points ahead of OpenAI's GPT-5.5-xhigh at 72.8% and significantly ahead of Claude Opus 4.6 at 70.9%.
What specific task does the BIRD benchmark test for SQL models?
The BIRD benchmark tests the ability to translate messy human questions into perfect SQL queries that execute against real-world databases with complex business logic. This is a challenging, practical task that requires understanding both natural language and intricate enterprise data structures.
How does Gemini-SQL2's performance compare to competitors like OpenAI and Claude?
Gemini-SQL2 leads all competitors with an 80.04% execution accuracy, maintaining a seven-point advantage over OpenAI's GPT-5.5-xhigh (72.8%) and a nine-point lead over Claude Opus 4.6 (70.9%). Other models from Databricks, AWS, Tencent, and Alibaba score significantly lower, putting them far behind in the rankings.
What engineering advantage does Gemini-SQL2 likely have over other models?
Gemini-SQL2's superior performance likely stems from specialized training on the complex rules and structures of actual enterprise data systems. This targeted training approach gives Google a clear technical advantage in handling the labyrinthine requirements of real-world database queries.
Why is Gemini-SQL2's BIRD benchmark performance significant for enterprise data work?
A tool that reliably translates plain-English queries into accurate SQL significantly reduces the tedium of data work, which is a major pain point in enterprises. Gemini-SQL2's strong performance establishes Google as having a clear technical claim to being that reliable tool, creating a competitive moat in the data query market.
Further Reading
- Google Releases Gemini-SQL2: Gemini 3.1 Pro Text-to-SQL Scores 80.04% on BIRD Single-Model Leaderboard — MarkTechPost
- How to get Gemini to deeply understand your database — Google Cloud Blog
- Google Research giới thiệu Gemini-SQL2: Mô hình Text-to-SQL mới ... — Facebook
- Gemini-SQL2 Hits 80% BIRD: Data Query Shift 2026 — Apple Podcasts
