Qwen-MT - A machine translation model launched by Ali Tongyi Qianwen
Qwen-MT is a machine translation model developed by the Alibaba Tongyi Qianwen team, based on the powerful Qwen3 architecture. The model supports high-quality translation between 92 languages, covering over 95% of the global population, and can meet diverse cross-language needs...
What is Qwen-MT?
Qwen-MT is a machine translation model developed by Alibaba's Tongyi Qianwen team, based on the powerful Qwen3 architecture. The model supports high-quality translation between 92 languages, covering over 95% of the global population and meeting diverse cross-language communication needs. Based on a lightweight MoE architecture, the model features low latency and low cost, with API call costs as low as $0.5 per million output tokens. The model supports features such as terminology intervention, domain suggestions, and translation memory, and can customize translation styles according to user needs. In both automatic and human evaluations, Qwen-MT demonstrates excellent translation quality and fluency, making it an ideal choice for achieving efficient and intelligent translation.
Main functions of Qwen-MT
- Multilingual supportIt supports mutual translation of 92 mainstream languages and dialects, covering more than 95% of the world's population and meeting a wide range of cross-language needs.
- Highly customizedIt offers terminology intervention, domain-specific hints, and translation memory features, allowing users to customize their translation style to adapt to complex professional scenarios.
- Low latency and low costBased 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 real-time requirements.
- High-quality translationIt performs excellently in both automatic and human evaluation, providing accurate and fluent translations and supporting translation tasks across multiple domains.
Qwen-MT Technical Principles
- powerful base modelBased on the Qwen3 architecture, it is trained with trillions of multilingual and translation data to enhance multilingual understanding capabilities.
- Reinforcement learning optimizationFurther improve translation accuracy and fluency by leveraging reinforcement learning techniques, and optimize model performance.
- Lightweight MoE architectureThe Mixture of Experts (MoE) architecture enables efficient computing and rapid response, reducing API call costs.
- Customization features implementationIt supports terminology intervention, domain-specific hints, and translation memory, ensuring that translation results meet specific needs through user-defined parameters and prompts.
Qwen-MT's project address
- Project official website: https://qwenlm.github.io/blog/qwen-mt/
- Experience the Demo Onlinehttps://huggingface.co/spaces/Qwen/Qwen3-MT-Demo
Application scenarios of Qwen-MT
- Cross-language content creation and publishingIt helps news media, social media, and content platforms quickly translate content into multiple languages, expanding their reach and user interaction.
- Enterprise internationalizationIt helps multinational companies, customer service, and business communication achieve multilingual support, accelerates internationalization, and improves customer satisfaction.
- EducationIt provides multilingual translation services for online education, academic research, and language learning, promoting the sharing of educational resources and international academic exchange.
- Law and Government AffairsUsed for multilingual translation of legal documents and government information to ensure legal accuracy and improve the internationalization of public services.
- Technology and DevelopmentIt supports software localization, API integration, and technical documentation translation, helping developers achieve efficient localization and technical communication.