Fannie Mae, Freddie Mac shares tumble after conservatorship comments
Investing.com -- Alibaba (NYSE:BABA) has introduced Qwen3-MT, an advanced machine translation model that supports 92 languages covering over 95% of the global population.
The new model builds upon the Qwen3 foundation and has been trained on trillions of multilingual and translation tokens to enhance its translation capabilities. The company has also incorporated reinforcement learning techniques to improve accuracy and fluency.
Qwen3-MT offers several key features including high customizability with terminology intervention, domain prompts, and translation memory. It utilizes a lightweight Mixture of Experts (MoE) architecture to deliver faster response times and reduced API costs, which can be as low as $0.5 per million output tokens.
According to Alibaba’s evaluations, Qwen-MT outperforms similarly sized models including GPT-4.1-mini, Gemini-2.5-Flash, and Qwen3-8B on multi-domain translation benchmarks. The company claims it maintains competitive translation quality compared to larger models like GPT-4.1, Gemini-2.5-Pro, and Qwen3-235B-A22B, while offering faster processing.
Human evaluation involving professional translators across ten major languages showed Qwen-MT achieved superior performance metrics in both acceptance rates and excellence rates.
The model supports languages from multiple language families including Indo-European, Sino-Tibetan, Afro-Asiatic, Austronesian, Dravidian, Turkic, Tai-Kadai, Uralic, Austroasiatic, and others.
Users can access Qwen-MT through the Qwen API, with the ability to customize translations for specific domains and terminology requirements.
This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.