Tencent Unveils 440MB AI Model for Offline Translation

Global AI Watch··3 min read·The Decoder
Tencent Unveils 440MB AI Model for Offline Translation

Key Takeaways

  • 1Tencent launched Hy-MT1.5 model for offline translations.
  • 2Model uses aggressive compression for efficiency and speed.
  • 3Enhances local AI capabilities, reducing dependence on cloud services.

Tencent has introduced an open-source AI translation model, Hy-MT1.5-1.8B-1.25bit, designed for offline use on smartphones. The model, weighing only 440 MB, can translate 33 languages and five dialects, achieving performance comparable to larger models like Qwen3-32B. This breakthrough was made possible through aggressive compression techniques, reducing the model size from 3.3 GB to 440 MB without compromising translation quality.

This development not only positions Tencent competitively against other translation services such as Google Translate but also highlights a strategic shift towards enhancing local AI capabilities. By enabling offline functionality, the Hy-MT1.5 model minimizes the need for internet connectivity, thereby reducing dependency on external cloud processing and improving data sovereignty for users. Such advancements in local AI not only cater to user convenience but also align with increasing demands for data privacy and efficiency in translation technology.

Related Sovereign AI Articles

Explore Trackers