Qwen (Tongyi Qianwen) is a comprehensive family of large language models developed by Alibaba Cloud that delivers state-of-the-art performance across text, code, vision, and audio tasks in both English and Chinese — available in sizes from 0.5B to 110B parameters with open weights, strong multilingual capabilities, dedicated coding variants (Qwen-Coder), vision-language models (Qwen-VL), and math-specialized versions (Qwen-Math) that make it one of the most versatile open-source model families available.
What Is Qwen?
- Definition: A series of transformer-based language models from Alibaba Cloud's Tongyi Lab — trained on multilingual data with particular strength in English and Chinese, released with open weights under permissive licenses (Apache 2.0 for most variants).
- Model Family: Qwen is not a single model but a comprehensive ecosystem — base models, chat models, coding models, vision-language models, math models, and audio models, each available in multiple sizes.
- Multilingual Strength: Trained on a diverse multilingual corpus with emphasis on English and Chinese — Qwen models consistently rank among the top performers on both English (MMLU, HumanEval) and Chinese (C-Eval, CMMLU) benchmarks.
- Size Range: 0.5B, 1.8B, 4B, 7B, 14B, 32B, 72B, and 110B parameter variants — the smaller models (0.5B, 1.8B) are specifically optimized for mobile and edge deployment.
Qwen Model Variants
| Variant | Focus | Sizes | Key Strength |
|---|---|---|---|
| Qwen2.5 | General purpose | 0.5B-72B | Balanced performance |
| Qwen2.5-Coder | Code generation | 1.5B-32B | Top open-source coding model |
| Qwen-VL | Vision-language | 7B-72B | Image understanding + OCR |
| Qwen2.5-Math | Mathematical reasoning | 1.5B-72B | Step-by-step math solving |
| Qwen-Audio | Audio understanding | 7B | Speech + sound recognition |
| Qwen2.5-Instruct | Chat/instruction | All sizes | Instruction following |
Why Qwen Matters
- Coding Excellence: Qwen-Coder models consistently rank among the best open-source coding models — competitive with or exceeding CodeLlama and DeepSeek-Coder on HumanEval, MBPP, and MultiPL-E benchmarks.
- Edge Deployment: The 0.5B and 1.8B models are specifically designed for mobile phones and IoT devices — small enough to run on-device while maintaining useful capabilities.
- Vision-Language: Qwen-VL handles image understanding, OCR, document parsing, and visual question answering — one of the strongest open-source VLMs available.
- Commercial License: Most Qwen variants are released under Apache 2.0 — fully permissive for commercial use without restrictions.
Qwen is the most comprehensive open-source model family from the Chinese AI ecosystem — providing state-of-the-art performance across text, code, vision, math, and audio in both English and Chinese, with sizes ranging from edge-deployable 0.5B to frontier-class 110B parameters under permissive open-source licenses.
Explore 500+ Semiconductor & AI Topics
From EUV lithography to CUDA optimization — search the full knowledge base or chat with our AI assistant.