According to AIbase, Tencent has officially released version 1.5 of its HY-MT translation model and announced that it will be open-sourced. This is a powerful solution supporting bidirectional translation among 33 languages, designed to provide excellent translation capabilities for mobile devices and efficient edge deployment.

QQ20251230-153609.png

The newly released model comes in two versions: Tencent-HY-MT1.5-1.8B and Tencent-HY-MT1.5-7B. The 1.8B version has been specially optimized, requiring only 1GB of memory to achieve offline real-time translation on consumer devices such as smartphones after quantization. Its inference speed is extremely fast, processing 50 tokens in just 0.18 seconds, far exceeding mainstream commercial translation APIs, making it highly practical in real-time translation scenarios.

In terms of performance, the HY-MT 1.5 model performs outstandingly on authoritative WMT25 and Flores200 test sets. Notably, the lightweight 1.8B model has reached the 90th percentile level of Google's large-scale closed-source model Gemini-3.0-Pro in translation quality and response efficiency, achieving an excellent balance between the two. The 7B version significantly improves translation accuracy compared to the previous version and effectively reduces unnecessary annotations and language mixing issues in the translated text.

The model not only leads in technology but also offers comprehensive features. It allows users to customize terminology libraries based on industry characteristics, ensuring accuracy and consistency in professional field translations. Additionally, its excellent context understanding capability ensures coherence and fluency in translating long texts and multi-turn dialogues, significantly enhancing user experience.

Tencent's HY-MT model has won multiple awards in international machine translation competitions and has been applied in core business products such as Tencent Meeting and Enterprise WeChat, proving its strong competitiveness.

  • HY-MT official website: https://hunyuan.tencent.com/modelSquare/home/list

  • Github link: https://github.com/Tencent-Hunyuan/HY-MT

  • HuggingFace link: https://huggingface.co/collections/tencent/hy-mt15