Google Releases Gemini-SQL2: Gemini 3.1 Pro Text-to-SQL Scores 80.04% on BIRD Single-Model Leaderboard

Google Research group has introduced the launch of Gemini-SQL2 on X. They described this method as a breakthrough text-to-SQL functionality powered by Gemini 3.1 Pro. Gemini-SQL2 posted 80.04% execution accuracy on the BIRD Text-to-SQL Leaderboard (Single Model). Google’s chart locations it above its personal Gemini-SQL, the prior high entry. The metric measures whether or not generated SQL runs and returns right outcomes, not whether or not it appears legitimate.

Gemini-SQL2

Gemini-SQL2 is a text-to-SQL functionality, not a standalone basis mannequin launch. It interprets pure language questions into what Google calls ‘execution-ready SQL queries.’ The functionality is constructed on Gemini 3.1 Pro.

Per the announcement on X, “knowledge subtlety & advanced enterprise contexts make producing correct SQL from pure language notoriously exhausting.” The X Post additionally said that “improved SQL understanding can elevate pure language abilities throughout Google’s knowledge providers.” That factors towards integration targets like BigQuery Studio, AlloyDB AI, and Cloud SQL Studio, which already ship Gemini-based SQL technology. Google has not but confirmed which merchandise will obtain Gemini-SQL2.

Benchmarks

BIRD (BIg Bench for LaRge-scale Database Grounded Text-to-SQL Evaluation) is an business normal for this job. It accommodates 12,751 question-SQL pairs throughout 95 databases spanning 37 skilled domains, totaling 33.4GB. The databases embody soiled values and require exterior information grounding, not like older benchmarks akin to Spider.

BIRD measures execution accuracy (EX): the generated SQL should run and return outcomes matching the gold question. Google said this straight. “Per the BIRD benchmark, which measures execution-verified accuracy, GeminiSQL-2’s SQL doesn’t simply look proper, it additionally runs efficiently.”

The Single Trained Model Track restricts the preprocessing, retrieval, and agentic frameworks that ensembles use to spice up scores. It measures the mannequin’s core text-to-SQL skill. Google Cloud’s prior document on this observe, reported November 15, 2025, was 76.13. Google benchmarks human efficiency at 92.96, leaving a 12.92-point hole from 80.04.

How the Leaderboard Stacks Up

Google’s chart, on X put up, reveals Gemini-SQL2 forward of eight named rivals, together with a number of unlabeled factors. Only 80.04% is said as textual content. The values under are learn from the chart’s place and are approximate; dates mirror every level’s horizontal placement.

Two patterns are seen. Google now holds the highest two named positions, Gemini-SQL2 and Gemini-SQL. Several specialised 32B SQL fashions additionally sit above some basic frontier fashions on this chart.

Use Cases with Examples

• Self-service analytics: A income supervisor asks for month-to-month recurring income by area, for accounts that churned inside 90 days of improve. This wants joins, window logic, and date arithmetic. Execution-verified technology catches SQL that runs however returns flawed rows.
• Data engineering drafts: Devs can draft BigQuery transformations from English, then evaluate somewhat than write from scratch. Google’s November 2025 work recognized schema understanding because the exhausting half. Higher BIRD scores mirror higher dealing with of ambiguous columns and messy values.
• Embedded “ask your knowledge” options: SaaS groups including natural-language question interfaces nonetheless want human evaluate at 80% accuracy. One in 5 queries may be flawed. The rating units expectations, not a elimination of evaluate.

Gemini-SQL2 Launch: Community Reception Dashboard

“+
“

“+
“

“;
field.appendChild(row);
});
perform fmt(n,f){
if(f===”okay”&&n>=1000){return (n/1000).toFixed(n>=100000?1:1).substitute(/.0$/,””)+”Ok”;}
return n.toLocaleString();
}
perform animate(){
Array.prototype.forEach.name(doc.querySelectorAll(“#mtp-gsql2-dash .dd-bar-fill”),perform(b){
b.type.setProperty(“width”,b.getAttribute(“data-w”)+”%”,”necessary”);
});
Array.prototype.forEach.name(doc.querySelectorAll(“#mtp-gsql2-dash .dd-num”),perform(el){
var finish=parseInt(el.getAttribute(“data-count”),10),f=el.getAttribute(“data-fmt”),
begin=null,dur=1100;
perform step(ts){if(!begin)begin=ts;var p=Math.min((ts-start)/dur,1);
el.textContent=fmt(Math.spherical(finish*(p*(2-p))),f);
if(p