It outperforms general reasoning models on complex database schemas.
Gemini is the new Prisma
🚀 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. 🧵👇
Positive users call Gemini-SQL2 cool and important for BigQuery relevance while negative users dismiss the SOTA claims due to benchmark leakage, misleading charts, and risks of hallucinations or failures in real production databases.
Text-to-SQL might sound like a solved problem.
Far from it. Data gets messy and complex really fast in the real world. Strong reasoning models are great, but nothing beats a custom model at this stuff. Gemini-SQL2 looks very strong here. BIRD is a tough benchmark. I suspect there are plenty of opportunities like this in KBs, search, graph databases, etc.
🚀 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. 🧵👇
@theo Letting an LLM generate sql still feels very risky. Using prisma without proper db knowledge already allows you to shoot in your own foot, let alone full on generating queries and migrations.
@theo what even is prisma
@theo
@julienzeroshot @theo Not really. Just use a db user/connection string that only has read access, then all you have to worry about is non performant queries
@theo Holy chart crime
@theo sql is already text...
@theo
@theo all fun and games until this thing hallucinates in your prod DB and blames 'natural language ambiguity'
@theo I think this is important given that Google has BigQuery which is a big player in data
@gilrdb @theo ORM
@theo I am genuinely curious who asked for this.
Like i think I want a non deterministic way to manage my database
@theo That' cool . interesting .
@julienzeroshot @theo This is not for production business logic but more for data retrieval for analytics/insights
@theo Downgrade
@theo lol savage
@theo altimate-code ftw
@theo Wtf is even Prisma?
@theo It's it good for coding. How do you compare it with opus and 5.5?
@omarsar0 benchmark numbers vs actually working on customer data is a gap most people dont talk about enough