Feyn AI Releases SQRL, a Text-to-SQL Model Family That Inspects the Database Before Writing a Query

Feyn Labs has introduced SQRL, an innovative family of text-to-SQL models designed to transform how artificial intelligence converts natural language into database queries. By implementing a unique database inspection methodology before query generation, SQRL represents a significant advancement in SQL generation technology. The flagship SQRL-35B model demonstrates exceptional performance, achieving 70.6% execution accuracy on the BIRD Dev benchmark—a competitive result that edges out Claude Opus—while offering scalable alternatives for various deployment scenarios.

The SQRL system operates through a distinctive two-stage process. Rather than immediately translating text to SQL, the model first conducts read-only database probes to understand the schema, structure, and relevant data relationships. This inspection-before-execution approach fundamentally improves query accuracy and relevance. The flagship SQRL-35B-A3B model achieves industry-leading performance metrics, while the family includes distilled versions—a 4B parameter model and a 9B parameter model—enabling self-hosted deployment without compromising accessibility or performance.

• Enhanced Query Accuracy: Database inspection before query generation reduces execution errors and improves contextual understanding of data relationships
• Scalability and Accessibility: Distilled 4B and 9B checkpoints make advanced SQL generation available for self-hosted and resource-constrained environments
• Competitive Performance: SQRL-35B's results rival leading proprietary models like Claude Opus, democratizing high-performance AI capabilities
• Enterprise Deployment: The self-hostable architecture addresses data privacy and security concerns critical for enterprise database applications
• Development Efficiency: Reduces the need for manual query optimization and debugging in data analytics workflows

The release of SQRL addresses a critical gap in AI-assisted database interaction, where accuracy directly impacts business intelligence, reporting, and data analysis outcomes. By combining competitive performance with practical deployment flexibility, SQRL enables organizations to leverage advanced SQL generation without vendor lock-in or prohibitive computational costs. This development signals a broader trend toward specialized, database-aware AI models that prioritize accuracy through architectural innovation rather than pure scale.