This repository is the official home of Qwen, Alibaba Cloud’s family of open large language models (known in Chinese as Tongyi Qianwen). It has collected nearly 22,000 stars on GitHub and hosts the original Qwen chat and pretrained model series under the Apache-2.0 license.
The Qwen repo contains the first generation of Alibaba’s open-weight language models, spanning pretrained base models and instruction-tuned chat models in a range of sizes for different hardware budgets. The models were notable for strong bilingual ability in Chinese and English, competitive benchmark results against much larger systems, and permissive open licensing that made them a popular foundation for research and commercial projects alike. The repository includes model weights links, inference and fine-tuning code, and documentation for deploying the models locally or in the cloud.
The model lineup in this repository spans from compact models small enough for laptops and edge devices up to a 72-billion-parameter flagship, with chat versions tuned for instruction following, dialogue, and tool use. The Qwen family quickly spawned specialized descendants for code and mathematics, and its releases helped establish a higher bar for open bilingual models at a time when most strong open weights were English-centric. While Alibaba has since published newer generations under their own repositories, this original repo remains the canonical starting point for the first Qwen series and its extensive downstream ecosystem of fine-tunes and tools.
The models and code in this repository are free and open source. Anyone can download, run, and fine-tune them without paying Alibaba, subject to the repository license terms.
This Qwen repository is for researchers and developers who want a proven open bilingual LLM family to build on, especially for Chinese-language applications, fine-tuning experiments, and self-hosted deployments.