Qwen

Definition

Qwen is a family of large language models developed by Alibaba Cloud, released across multiple open-weight and proprietary versions spanning general chat, coding, reasoning, and multimodal variants.

Overview

Qwen (also styled Tongyi Qianwen) is Alibaba's family of large language models, first released in 2023 and iterated through successive generations (Qwen, Qwen1.5, Qwen2, Qwen2.5, and Qwen3). Many Qwen models are released as open-weight models, meaning their trained parameters are published for developers to download, run locally, and fine-tune, which has made Qwen one of the most widely adopted open model families outside of Meta's Llama series, particularly popular on platforms like Hugging Face. The Qwen family spans a wide range of sizes, from small models suited to on-device or edge deployment up to large dense and mixture-of-experts models competitive with leading proprietary systems on many benchmarks. Alibaba also ships specialized variants such as Qwen-Coder for programming tasks and Qwen-VL for vision-language understanding, alongside the general-purpose chat models. Qwen is trained using large-scale pretraining on multilingual text (with particular strength in Chinese and English) followed by instruction tuning and alignment, and it is used both directly through Alibaba Cloud's API and by third parties who self-host the open-weight checkpoints for cost control, privacy, or customization via fine-tuning.

Key Features

• Family of models spanning multiple generations (Qwen through Qwen3)
• Many variants released as open-weight models on platforms like Hugging Face
• Sizes ranging from small edge-friendly models to large flagship models
• Specialized variants for coding (Qwen-Coder) and vision-language tasks (Qwen-VL)
• Strong multilingual performance, especially Chinese and English
• Some flagship variants use a mixture-of-experts architecture
• Accessible via Alibaba Cloud API or self-hosted from published weights

Use Cases

Frequently Asked Questions

Related in AI Models