LuthorSQL: A One-Pass Text-to-SQL Model as System 1 for a Slower LLM Agent

arXiv cs.CL Papers

Summary

LuthorSQL presents a non-autoregressive 7B text-to-SQL model acting as a fast System 1 gate that escalates uncertain queries to a slower LLM agent, achieving 68.03% execution accuracy on BIRD while answering nearly half of queries locally and cutting API costs by 33.3% when cascaded with Claude Opus 5.5.

arXiv:2611.17780v1 Announce Type: new Abstract: Frontier LLM agents that query databases with tools are among the most accurate text-to-SQL systems, but per-query calls to a hosted model are slow, costly, and expose sensitive organizational queries to external APIs. For example, a Claude Opus 5.5 agent serving 10,000 queries a day costs over $70,000 a year at list price. To what extent can local hardware already owned by an organization offload this burden without compromising the agent's accuracy? We present LuthorSQL, a non-autoregressive 7B text-to-SQL model that serves as a rapid System 1 gate for a slower LLM agent (System 2), escalating queries only when uncertain. LuthorSQL makes a single forward pass to predict slot scores over a bounded grammar for query plans, compiles and runs up to 32 candidate plans, and chooses an answer using a trained validator. On its own, LuthorSQL reaches 68.03% execution accuracy on BIRD's hidden test set at 0.84 s per question on a single GPU, maintains its accuracy in 9 GB with 8-bit quantized weights, and transfers zero-shot to Snowflake, BigQuery, MySQL, PostgreSQL, and DuckDB within 3.1 points of its native SQLite accuracy. Combined with Claude Opus 5.5 in a pre-registered confidence cascade over 227 questions on databases published after Opus's training cutoff, LuthorSQL answers nearly half of all questions locally and cuts API cost by 33.3% at the same accuracy. Finally, contamination probing shows that Opus reproduces 48-92% of public benchmark gold queries verbatim without seeing the schema; however, its tool-assisted accuracy holds on renamed schemas and reaches 95.6-97.4% on post-cutoff databases. Code and data: https://github.com/IAMIbrahimmemon/luthorsql
Original Article
View Cached Full Text

Cached at: 11/16/26, 08:21 AM

# LuthorSQL: A One-Pass Text-to-SQL Model as System 1 for a Slower LLM Agent
Source: [https://arxiv.org/abs/2611.17780](https://arxiv.org/abs/2611.17780)
[View PDF](https://arxiv.org/pdf/2611.17780)

> Abstract:Frontier LLM agents that query databases with tools are among the most accurate text\-to\-SQL systems, but per\-query calls to a hosted model are slow, costly, and expose sensitive organizational queries to external APIs\. For example, a Claude Opus 5\.5 agent serving 10,000 queries a day costs over $70,000 a year at list price\. To what extent can local hardware already owned by an organization offload this burden without compromising the agent's accuracy? We present LuthorSQL, a non\-autoregressive 7B text\-to\-SQL model that serves as a rapid System 1 gate for a slower LLM agent \(System 2\), escalating queries only when uncertain\. LuthorSQL makes a single forward pass to predict slot scores over a bounded grammar for query plans, compiles and runs up to 32 candidate plans, and chooses an answer using a trained validator\. On its own, LuthorSQL reaches 68\.03% execution accuracy on BIRD's hidden test set at 0\.84 s per question on a single GPU, maintains its accuracy in 9 GB with 8\-bit quantized weights, and transfers zero\-shot to Snowflake, BigQuery, MySQL, PostgreSQL, and DuckDB within 3\.1 points of its native SQLite accuracy\. Combined with Claude Opus 5\.5 in a pre\-registered confidence cascade over 227 questions on databases published after Opus's training cutoff, LuthorSQL answers nearly half of all questions locally and cuts API cost by 33\.3% at the same accuracy\. Finally, contamination probing shows that Opus reproduces 48\-92% of public benchmark gold queries verbatim without seeing the schema; however, its tool\-assisted accuracy holds on renamed schemas and reaches 95\.6\-97\.4% on post\-cutoff databases\. Code and data:[this https URL](https://github.com/IAMIbrahimmemon/luthorsql)

## Submission history

From: Ibrahim Memon \[[view email](https://arxiv.org/show-email/07e255ff/2611.17780)\] **\[v1\]**Fri, 13 Nov 2026 05:44:08 UTC \(66 KB\)

Similar Articles

DualSQL: Text-to-SQL with Multi-Agent Reinforcement Learning

arXiv cs.CL

DualSQL proposes a multi-agent reinforcement learning framework for Text-to-SQL, using a single shared model to jointly optimize schema linking and SQL generation, achieving state-of-the-art accuracy with smaller model sizes.