Tencent Hy-MT2: 440MB Open-Source Model Beats Microsoft Translation API

Tencent Hy-MT2: 440MB Open-Source Model Beats Microsoft Translation API

Tencent's Hunyuan team released Hy-MT2, an open-source multilingual translation model supporting 33 languages in three sizes: 1.8B, 7B, and 30B-A3B. The 7B and 30B variants achieve state-of-the-art among open-source translation models, while the 1.8B — compressed using Tencent's AngelSlim 1.25-bit quantization to 440MB — outperforms Microsoft's commercial translation API. The model is 1.5x faster than its predecessor, runs on mainstream mobile chips locally, and is available on HuggingFace, Modelscope, and GitHub under a fully open-source license.

Why It Matters

A 440MB model outperforming a commercial translation API at 33-language coverage sets a new bar for on-device multilingual deployment. Combined with mobile chip compatibility, Hy-MT2 has immediate implications for privacy-preserving translation in both consumer applications and enterprise workflows where data sovereignty is required.