chiprook
← AI
AIOctober 20, 2026, 09:46

Two prompt rules lift Oracle text-to-SQL from 35% to 64%

A developer ran the schemagate tool with the light Gemini 2.5 Flash model on UniQL's Oracle benchmark of 1,534 questions. Adding two Oracle rules to the prompt — no AS before table aliases and quoting reserved words DATE and NUMBER — raised first-try accuracy from 35.0% to 64.5%, above the best published 49.61% from Claude Sonnet 4.5.

Two prompt rules lift Oracle text-to-SQL from 35% to 64%
#Google#Gemini#Oracle#Anthropic
Read next
AI

Gemini 4 Argon agents beat Minecraft; MLExpert lifts MLE-Bench medal rate

AI

Gemini 4 Argon vs Opus 5.5 vs GPT-6 Astra: same prompts, real bills

AI

Google launches Playground to turn prompts into shareable browser games

AI

Google replaces Gemini Gems with Skills — reusable prompts in any chat