Spider

Spider is a large-scale cross-domain semantic parsing and text-to-SQL benchmark featuring 10,181 questions and 5,693 unique complex SQL queries across 200 databases covering 138 different domains. This comprehensive evaluation tests AI models' ability to understand natural language queries and generate accurate SQL code for complex database operations, requiring sophisticated semantic understanding and SQL generation capabilities.