Alibaba's large model team Qwen (Tongyi Qianwen) has released its latest multilingual translation model, Qwen3-MT, which supports 92 languages and covers more than 95% of the world's population, making it one of the most powerful general-purpose translation models at present.
1. Wide range of support and strong language capabilities
Qwen3-MT is trained on trillions of multilingual data, covering mainstream languages and a large number of minor languages, and is suitable for global products, cross-border content, and localization services.
2. Outstanding advantages of the model
- High translation quality: accurate and natural, close to the human level
- Support customization: support terminology, domain control and translation memory (TM)
- fast speed and low price : Adopts MoE architecture, the cost is only $0.5 / million tokens
- can be deployed on a large scale: low latency, high concurrency, suitable for real-time translation needs
3. Multi-platform open access
- to Hugging Face Demonstration:
- https://huggingface.co/spaces/Qwen/Qwen3-MT-Demo
- ModelScope demo:
- https://modelscope.cn/studios/Qwen/Qwen3-MT-demo
- API documentation (Alibaba Cloud):
- https://alibabacloud.com/help/en/model-studio/translation-abilities
- official blog:
- https://qwenlm.github.io/blog/qwen-mt/
4. Summary
Qwen3-MT is a high-quality translation model suitable for the development of overseas products, multilingual content platforms, and AI tools, with outstanding performance and low threshold for use, making it worthy of attention and access.
