Qwen-MT machine translation model is officially released. Qwen-MT is based on the powerful Qwen3 model, trained on a large-scale multilingual and translation dataset, combined with reinforcement learning technology, significantly improving the accuracy and fluency of translation results. Developers can experience its fast and accurate translation capabilities directly through the Qwen API (qwen-mt-turbo).
The core highlights of Qwen-MT include supporting bidirectional translation of 92 languages, covering more than 95% of the global population, meeting extensive language communication needs. In addition, the model also has high controllability, providing professional translation functions such as terminology intervention, domain prompts, and memory libraries, and supports users to customize prompts, effectively enhancing the translation performance of the model in complex, professional, or specific application scenarios. In terms of performance, Qwen-MT adopts a lightweight MoE (Mixture of Experts) architecture, achieving faster response speed and lower API call cost (as low as 2 yuan per million output tokens), making it more suitable for high concurrency and real-time applications.
In automatic evaluation, Qwen-MT significantly outperforms models of similar scale, such as GPT-4.1-mini, Gemini-2.5-Flash, and Qwen3-8B, in multiple fields of Chinese-English, English-German translation, and WMT24 multilingual translation tasks. Even compared with top-tier large models like GPT-4.1, Gemini-2.5-Pro, and Qwen3-235B, the translation effect of Qwen-MT remains no less impressive, thanks to its lightweight model architecture design that provides a fast translation experience.
To more accurately evaluate translation quality, Qwen-MT also conducted manual evaluation. For major languages such as Chinese, English, Japanese, Korean, Thai, Arabic, Italian, Russian, Spanish, and French, manual evaluation was carried out based on real scenario translation data. Each test sample was independently scored by three professional translators and cross-verified to ensure the objectivity and reliability of the evaluation results. In terms of pass rate and excellent rate, Qwen-MT shows significant advantages, demonstrating its outstanding translation capabilities in practical applications.
The release of Qwen-MT marks a significant advancement in machine translation technology. It not only supports bidirectional translation of multiple languages, but also has high controllability, low latency, and low cost, providing global users with a smarter, more flexible, and more efficient translation experience. In the future, Qwen-MT will continue to improve the accuracy and naturalness of translation, expand coverage of more language pairs, striving to break down language barriers and make communication without language obstacles.
Bailian: https://bailian.console.aliyun.com/?tab=model#/model-market/detail/qwen-mt-turbo