XiYan SQL
A MULTI-GENERATOR ENSEMBLE FRAMEWORK FOR NATURAL LANGUAGE TO SQL
Ask AI about XiYan SQL
Powered by Claude Β· Grounded in docs
I know everything about XiYan SQL. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Reviews
Documentation
π€ ζθ¨GBI | π€ XiYanSQL Model | π» M-Schema | π Arxiv| π Preview| π PapersWithCode
δΈζη | θ±ζη
XiYan-SQL: A Multi-Generator Ensemble Framework for Text-to-SQL
Here is the new official Alibaba repository for XiYan-SQL. We will currently maintain synchronization between these two addresses.
Newsπ₯
-
Nov. 21, 2025π New SOTA on BIRD-CRITIC-Open: XiYan-SQL-CRITIC technique has achieved an impressive 44.37% success rate on BIRD-CRITIC-Open, a highly challenging multi-dialect benchmark, securing the top position with SOTA performance! -
Oct. 30, 2025π We are excited to release the XiYan-SQL training framework XiYan-SQLTraining! This framework is primarily designed for the training of SQL/general LLMs and includes capabilities such as SQL data processing, model training, and evaluation as proposed by XiYan. We will continue to enhance the framework in the future. -
Oct. 20, 2025π New SOTA on BIRD-CRITIC: XiYan-SQL-CRITIC technique has achieved a remarkable 44.53% success rate on the BIRD-CRITIC-PG benchmark, securing the top position with SOTA performance! Additionally, it recorded an impressive 48.5% success rate on the BIRD-CRITIC-Flash benchmark, also establishing a new SOTA performance. -
Oct. 20, 2025π The training framework of XiYan-SQL, XiYan-SQLTraining, will soon be released as open source in the official Alibaba repository of XiYan-SQL. Stay tuned! -
...
-
May. 22, 2025π New SOTA on BIRD-CRITIC-Flash: XiYanSQL-CRITIC algorithm achieves a 41% Pass Rate score on the BIRD-CRITIC-Flash benchmark, setting a new SOTA performance. -
Apr. 29, 2025π We are excited to update our new XiYanSQL-QwenCoder-2504 series models. This version includes significant optimizations over the previous version, achieving new SOTA performance for single models. We welcome everyone to try it out. -
Mar. 21, 2025π Want a high security data access? XiYan-MCP-server now supports the local mode, which runs XiYanSQL-QwenCoder-3B on your own PC/Mac. -
Mar. 20, 2025π We are excited to announce that XiYanSQL-QwenCoder-32B can be used directly through ModelScope API. For details, see XiYanSQL-QwenCoder-32B-2412 -
Mar. 13, 2025πWe release a MCP server of XiyanSQL: XiYan-MCP-server. -
Mar. 10, 2025πWe have fully released our XiYanSQL-QwenCoder series models on the HuggingFace platform: HuggingFace. -
Mar. 04, 2025πWe release the method and source code of automatically generating database description for Text-to-SQL: Database Description Generation. -
Feb. 26, 2025πWe are excited to open source the XiYanSQL-QwenCoder series model, dedicated to advancing the development of LLMs in the text-to-SQL domain. As of now, XiYanSQL-QwenCoder covers four mainstream model sizes: 3B, 7B, 14B, and 32B parameters, to meet the needs of different developers. -
Jan. 22, 2025πWe release XiYanSQL-QwenCoder-32B and simultaneously open source the model weights. -
Jan. 09, 2025πXiYanSQL-QwenCoder-32B: XiYanSQL-QwenCoder-32B achieves an EX score of 69.03% on the BIRD test set, setting a new SOTA under only a single fine-tuned model. -
Dec. 17, 2024πNew SOTA on Bird: XiYan-SQL reaches the top of Bird leaderboard with an EX score of 75.63%, outperforming the second place by 0.84 pt. It also achieves a new SOTA with an R-VES score of 71.41%. -
Dec. 13, 2024We release the model and source code of DateResolver. -
Dec. 12, 2024Try our model: ModelScope
π Join Our Team
Our team is hiring interns and class-of-2027 graduates in Deep Research, LLM post-training, AI Agents, and NL2SQL. With solid technical expertise and cutting-edge algorithm R&D projects, we encourage academic research and top-tier conference publications. Candidates passionate about LLMs are welcome to send their resumes to zhencang.lyf@alibaba-inc.com.
Introduction in Short.
XiYan-SQL is an innovative framework for LLM based Text-to-SQL.
It contains:
-
XiYanSQL-QwenCoders Multiple different sizes of XiYanSQL models for SQL generation.
-
XiYan-SQLTraining A post-training framework specifically designed for the Text-to-SQL task developed by XiYan.
-
XiYan-mcp A MCP server that enables natural language queries to databases powered by XiYan-SQL, SOTA of text-to-sql on open benchmarks.
-
M-schema a semi-structured schema representation method.
-
Database Description Generation a method and corresponding code for automatic description generation for Text-to-SQL.
-
DateResolver a date understanding and reasoning enhanced model, major for Chinese.
-
MoMQ a multi-dialects Text-to-SQL MoE model based on Qwen.
-
...
π We welcome everyone to contribute to the XiYanSQL project !!!
Full Intro.
To tackle the challenges of large language model performance in natural language to SQL tasks, we introduce XiYan-SQL, an innovative framework that employs a multi-generator ensemble strategy to improve candidate generation. We introduce M-Schema, a semi-structured schema representation method designed to enhance the understanding of database structures. To enhance the quality and diversity of generated candidate SQL queries, XiYan-SQL integrates the significant potential of in-context learning (ICL) with the precise control of supervised fine-tuning. On one hand, we propose a series of training strategies to fine-tune models to generate high-quality candidates with diverse preferences. On the other hand, we implement the ICL approach with an example selection method based on named entity recognition to prevent overemphasis on entities. The refiner optimizes each candidate by correcting logical or syntactical errors. To address the challenge of identifying the best candidate, we fine-tune a selection model to distinguish nuances of candidate SQL queries. The experimental results on multiple dialect datasets demonstrate the robustness of XiYan-SQL in addressing challenges across different scenarios. Overall, our proposed XiYan-SQL achieves the state-of-the-art execution accuracy of 75.63% on Bird test, 89.65% on the Spider test set, 69.86% on SQL-Eval, 41.20% on NL2GQL. The proposed framework not only enhances the quality and diversity of SQL queries but also outperforms previous methods.
Timeline
The major events.
| Date | Event |
|---|---|
| 2025-11 | XiYan-SQL-CRITIC technique has achieved an impressive 44.37% success rate on BIRD-CRITIC-Open, a highly challenging real-world multi-dialect benchmark, securing the top position with SOTA performance! |
| 2025-10 | We are excited to release the XiYan-SQL training framework XiYan-SQLTraining !!! This framework is primarily designed for the training of SQL/general LLMs and includes capabilities such as SQL data processing, model training, and evaluation as proposed by XiYan. We will continue to enhance the framework in the future. |
| 2025-10 | XiYan-SQL-CRITIC technique has achieved a remarkable 44.53% success rate on the BIRD-CRITIC-PG benchmark, securing the top position with SOTA performance! Additionally, it recorded an impressive 48.5% success rate on the BIRD-CRITIC-Flash benchmark, also establishing a new SOTA performance. |
| 2025-09 | The download count for the XiYanSQL-QwenCoder series models on ModelScope has exceeded 100k , making it the most influential SQL model in the field. |
| 2025-05 | XiYanSQL-CRITIC algorithm achieves a 41% Pass Rate score on the BIRD-CRITIC-Flash benchmark, setting a new SOTA performance. |
| 2025-04 | We have released version 2504 of the XiYanSQL-QwenCoder series models, which features enhanced performance compared to the previous version. It still includes four different parameter sizes: 3B, 7B, 14B, and 32B. We encourage everyone to utilize these models. |
| 2025-02 | We have released the XiYanSQL-QwenCoder series model, which includes four different sizes: 3B, 7B, 14B, and 32B parameters, to meet the needs of different developers. |
| XiYanSQL-QwenCoder-32B has been released | |
| 2025-01 | XiYanSQL-QwenCoder-32B achieves an EX score of 69.03% on BIRD test, new SOTA using only single fine-tuned model |
| 2024-12 | Reaching the top of Bird leaderboard with an EX score of 75.63% and R-VES of 71.41(new SOTA) |
| 2024-11 | Proposing XiYanSQL technology A Multi-Generator Ensemble Framework for Text-to-SQL |
| Achieving 41.20% on NL2GQL, and a competitive score of 72.23% on Bird dev (bird) | |
| Achieving 89.65% on Spider test set (new SOTA), 69.86% on SQL-Eval (new SOTA) | |
| 2024-10 | Proposing an SQL MoE model MoMQ |
| 2024-09 | Proposing DateSolver module |
| 2024-05 | Proposing M-schema, involving ICL in SQL generation |
| Achieving 86.98% on Spider test set (SOTA 86.6%) |
Application
Welcome everyone to try the intelligent data querying solution based on XiYan-SQL, which is called XiYan GBI. We welcome any product experiences and suggestions for optimization.
For product introduction, please visit: https://help.aliyun.com/zh/model-studio/user-guide/brief-introduction-of-gbi-products
To try the product, please visit: https://bailian.console.aliyun.com/xiyan
Product DingTalk Group: 94725009401
Contact us:
If you are interested in our research or products, please feel free to contact us.
Contact Information:
Yifu Liu, zhencang.lyf@alibaba-inc.com
Join Our DingTalk Group
Others
Star History
Citation
If you find our work helpful, feel free to give us a cite.
@ARTICLE{XiYanSQL,
author={Liu, Yifu and Zhu, Yin and Gao, Yingqi and Luo, Zhiling and Li, Xiaoxia and Shi, Xiaorong and Hong, Yuntao and Gao, Jinyang and Li, Yu and Ding, Bolin and Zhou, Jingren},
journal={IEEE Transactions on Knowledge and Data Engineering},
title={XiYan-SQL: A Novel Multi-Generator Framework for Text-to-SQL},
year={2026},
volume={},
number={},
pages={1-14},
doi={10.1109/TKDE.2026.3657851}}
@article{xiyansql_pre,
title={A Preview of XiYan-SQL: A Multi-Generator Ensemble Framework for Text-to-SQL},
author={Yingqi Gao and Yifu Liu and Xiaoxia Li and Xiaorong Shi and Yin Zhu and Yiming Wang and Shiqi Li and Wei Li and Yuntao Hong and Zhiling Luo and Jinyang Gao and Liyu Mou and Yu Li},
year={2024},
journal={arXiv preprint arXiv:2411.08599},
url={https://arxiv.org/abs/2411.08599},
primaryClass={cs.AI}
}