Sqrl text-to-sql: Feyn AI Launches SQRL

Feyn AI revealed SQRL on July 19, 2026. This new lineup of advanced models aims to boost query accuracy by taking a close look at databases before generating SQL queries. This fresh approach helps tackle the typical issues that come with traditional text-to-SQL systems.

What sets the SQRL family apart is its ability not just to convert natural language into SQL but also to perform a detailed inspection of the database schema and content. By doing this preliminary analysis, SQRL aims to eliminate confusion and ensure that the SQL queries it generates match the actual data structure and values within the database. This means that SQRL models can craft more dependable queries, which ultimately enhances execution accuracy.

The standout model, SQRL-35B-A3B, has really impressed people with its performance, hitting a 70.6% execution accuracy on the BIRD Dev benchmark. This score puts it ahead of rivals like Claude Opus 4.6, which recorded a 68.77% accuracy under the same testing conditions. Other models in the SQRL lineup include SQRL-4B and SQRL-9B, all available on Hugging Face.

Even with these impressive abilities, some important details are still under wraps. For instance, Feyn AI hasn’t shared the specific pricing for the SQRL models, leaving potential users unsure about the financial commitment. They also haven’t confirmed the launch timeline or hardware specifications like RAM and processor details.

|———————|——————–|——————–|——————–|

| Execution Accuracy | N/A | N/A | 70.6% |

| Benchmark Used | N/A | N/A | BIRD Dev |

The importance of SQRL’s inspection ability can’t be overlooked. Traditional text-to-SQL systems often produce queries that look correct but result in errors, due to issues like ambiguous column names or incorrect table joins.

For example, a query might join two tables in a wrong way or filter for values that don’t exist, leading to misleading outputs. These problems can be tricky because they often don’t trigger errors, making them hard to catch just through execution.

The BIRD benchmark provides a deeper insight into these failures, helping to show how often models return incorrect results despite generating valid SQL.

Looking forward, Feyn AI stands to make a significant difference in how businesses engage with their data. As organizations lean more on data-driven decisions, the demand for accurate and context-aware SQL queries is rising. SQRL offers not just a solution but also sets a new standard for future advancements in text-to-SQL systems. For more details, check out VentureBeat AI.

If you’re thinking about adopting a text-to-SQL model, SQRL’s innovative approach is definitely worth considering. Its ability to inspect databases before generating queries suggests it can deliver more accurate results than earlier models. When choosing between SQRL models, think about your specific database needs and how complex your data interactions are.