What is Qwen-MT
Qwen-MT is a machine translation model launched by the Alibaba Tongyi Qianwen team, built on the powerful Qwen3 architecture. The model supports high-quality mutual translation across 92 languages, covering more than 95% of the global population and meeting diverse cross-lingual communication needs. The model is based on a lightweight MoE architecture, featuring low latency and low cost, with the API call cost per million output tokens as low as $0.5. The model supports features such as terminology intervention, domain prompts, and translation memory, and can customize translation style according to user needs. In both automatic and human evaluations, Qwen-MT demonstrates excellent translation quality and fluency, making it an ideal choice for efficient and intelligent translation.
Main Features of Qwen-MT
- Multilingual Support: Supports mutual translation across 92 mainstream languages and dialects, covering more than 95% of the global population and meeting a wide range of cross-lingual needs.
- High Customizability: Provides terminology intervention, domain prompts, and translation memory features, allowing users to customize translation style and adapt to complex professional scenarios.
- Low Latency and Low Cost: Based on a lightweight MoE architecture, it offers fast response times and low API call costs (as low as $0.5 per million output tokens), making it suitable for applications with high concurrency and high real-time requirements.
- High-Quality Translation: Performs excellently in both automatic and human evaluations, with accurate and fluent translation, and supports translation tasks in multiple domains.
Technical Principles of Qwen-MT
- Powerful Foundation Model: Based on the Qwen3 architecture and trained with trillion-level multilingual and translation data, enhancing multilingual understanding capabilities.
- Reinforcement Learning Optimization: Based on reinforcement learning technology, it further improves translation accuracy and language fluency, optimizing model performance.
- Lightweight MoE Architecture: Uses a Mixture of Experts (MoE) architecture to achieve efficient computation and fast response, reducing API call costs.
- Implementation of Customization Features: Supports terminology intervention, domain prompts, and translation memory, ensuring that translation results meet specific needs through user-defined parameters and prompts.
Project Address of Qwen-MT
- Project Official Website：https://qwenlm.github.io/blog/qwen-mt/
- Online Experience Demo：https://huggingface.co/spaces/Qwen/Qwen3-MT-Demo
Application Scenarios of Qwen-MT
- Cross-Lingual Content Creation and Publishing: Helps news media, social media, and content platforms quickly translate content into multiple languages, expanding reach and user interaction.
- Enterprise Internationalization: Helps multinational enterprises, customer service, and business communication achieve multilingual support, accelerating internationalization and improving customer satisfaction.
- Education Sector: Provides multilingual translation for online education, academic research, and language learning, promoting the sharing of educational resources and international academic exchange.
- Law and Government Affairs: Used for multilingual translation of legal documents and government information, ensuring legal accuracy and enhancing the internationalization level of public services.
- Technology and Development: Supports software localization, API integration, and technical documentation translation, helping developers achieve efficient localization and technical exchange.