1 related articles
A new open-source benchmark quantifies how a 4KB semantic layer boosts LLM Text-to-SQL accuracy across Claude and GPT models, validated with McNemar's test.