# Agentar-Scale-SQL: Advancing Text-to-SQL through Orchestrated Test-Time Scaling

**Authors:** Pengfei Wang, Baolin Sun, Xuemei Dong, Yaxun Dai, Hongwei Yuan, Mengdie Chu, Yingqi Gao, Xiang Qi, Peng Zhang, Ying Yan  
**arXiv:** [2509.24403](https://arxiv.org/abs/2509.24403)  
**Date:** 2025-09-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2509.24403v6/

## Key points

- Agentar-Scale-SQL is a cutting-edge framework developed by a team of researchers to enhance Text-to-SQL performance.
- The framework addresses the gap in existing methods compared to human experts on challenging benchmarks like BIRD.
- It introduces an Orchestrated Test-Time Scaling strategy combining Internal Scaling, Sequential Scaling, and Parallel Scaling perspectives.
- Agentar-Scale-SQL is adaptable to new databases and more powerful language models, offering flexibility for various contexts.
- Extensive experiments have shown that it achieves State-of-the-Art (SOTA) performance on the BIRD benchmark with an execution accuracy of 81.67% on the test set.
- The framework represents a significant advancement in Text-to-SQL methods by enhancing reasoning processes within models and setting a new standard for performance improvement in complex language understanding tasks.

## Summary

, , , , 

Agentar-Scale-SQL: A Comprehensive Framework for Enhancing Text-to-SQL Performance

 is a cutting-edge framework developed by a team of researchers including Pengfei Wang, Baolin Sun, Xuemei Dong, Yaxun Dai, Hongwei Yuan, Mengdie Chu, Yingqi Gao, Xiang Qi, Peng Zhang, and Ying Yan. This innovative framework aims to address the existing gap in  methods compared to human experts on challenging benchmarks like the . The current approaches in this field lack an organized strategy for  and often overlook the model's internal reasoning process. To overcome these limitations, Agentar-Scale-SQL introduces a novel approach that leverages scalable computation to enhance performance significantly. The framework implements an Orchestrated Test-Time Scaling strategy that combines three key perspectives: Internal Scaling through RL-enhanced Intrinsic Reasoning, Sequential Scaling via Iterative Refinement, and Parallel Scaling using Diverse Synthesis and Tournament Selection. By synergistically integrating these perspectives, Agentar-Scale-SQL offers a comprehensive solution for improving Text-to-SQL performance. One of the notable features of Agentar-Scale-SQL is its adaptability to new databases and more powerful language models. This general-purpose framework provides flexibility for researchers and practitioners to apply it across various contexts effectively. Extensive experiments conducted with Agentar-Scale-SQL have demonstrated its ability to achieve State-of-the-Art (SOTA) performance on the BIRD benchmark. With an impressive execution accuracy of 81.67% on the test set and securing the top position on the official leaderboard, Agentar-Scale-SQL showcases a promising path towards achieving human-level performance in Text-to-SQL tasks. In conclusion, Agentar-Scale-SQL represents a significant advancement in the field of Text-to-SQL methods by introducing a well-orchestrated approach to  that enhances  within models. With its innovative strategies and adaptable nature, this framework sets a new standard for performance improvement in complex language understanding tasks like SQL query generation from text inputs.