AI startup Dacube said on Monday it ranked No. 1 worldwide in the Lite category of Spider 2.0, an international text-to-SQL performance benchmark.

The result follows its No. 1 finish in the DBT (Data Build Tool) category in July last year. Spider 2.0 measures how accurately AI converts human language, or natural language, into the database language SQL. It is used as a yardstick to validate technical capabilities because it handles complex data structures similar to real corporate environments, drawing competition from global big tech companies and leading research institutions.

Dacube said it scored 65.81 points in the Lite category and received the highest technical score, beating Snowflake and others.

Dacube said it has focused investment on NL2SQL technology for about 4 years since its founding. The company said it has advanced its proprietary natural language processing engine and AI agent architecture and applied them to its services, while expanding the scope of application across finance, accounting and broader data platforms.

Dacube CTO Ha-jeong Kim (김하정) said it boosted training efficiency and processing speed by running AI agents in parallel that handle problem-solving, result verification and failure analysis. He said the result combined its understanding of large language models with its AI agent design capabilities.

Dacube CEO Ye-ji Yoon (윤예지) said the DBT category was a newly established track, while the Lite category has been a core category over the past 2 years in which global big tech companies and leading research institutions have competed. She said an architecture that automates the entire problem-solving process with AI agents made repeatable results possible.