88Z AI, 30 April — Tencent Hunyuan has open-sourced the Hy-MT1.5-1.8B-1.25bit model for offline on-device translation, supporting 33 languages and 1,056 translation directions. Compressed to just 440MB, it runs locally on mobile phones without an internet connection. Built on the 1.8B-parameter Hy-MT1.5, it uses Sherry sparse quantization technology (accepted by ACL 2026) and delivers translation quality surpassing mainstream systems like Google Translate.
At a glance
- Model compressed to 440MB, supports offline local operation for 33 languages
- Uses ACL 2026-accepted technology, outperforming mainstream translation systems
The update
Tencent Hunyuan has officially open-sourced Hy-MT1.5-1.8B-1.25bit, a large language model designed for offline translation on mobile devices. Covering 33 languages and supporting up to 1,056 translation directions, the model is compressed to under 440MB, enabling efficient local operation on mobile devices without an internet connection. Built on the 1.8B-parameter Hy-MT1.5 architecture, the model innovatively applies Sherry sparse quantization technology, which was accepted by ACL 2026. Thanks to this technical breakthrough, the model maintains a lightweight footprint while delivering translation accuracy and fluency that surpass mainstream systems such as Google Translate, providing high-quality offline translation experiences for mobile users.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 腾讯混元Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.