Tencent Hy-MT2 multilingual translation models released with AngelSlim quantization
Deploy Hy‑MT2 models for high‑quality multilingual translation and leverage AngelSlim quantization for efficient on‑device inference.
Integrate Hy‑MT2 models into translation pipelines and test AngelSlim quantized 1.8B for on‑device use.
Summary
Tencent released the Hy‑MT2 family of multilingual translation models, available in 1.8B, 7B, and 30B‑A3B (MoE) sizes. The models support translation among 33 languages and follow translation instructions in multiple languages. AngelSlim 1.25‑bit extreme quantization reduces the 1.8B model to 440 MB and improves inference speed by 1.5×. Benchmarks show that the 7B and 30B‑A3B models outperform open‑source models like DeepSeek‑V4‑Pro and Kimi K2.6, while the 1.8B model surpasses commercial APIs from Microsoft and Doubao. Tencent also released the IFMTBench benchmark and a Hy‑MT2‑Translator Skill for easy integration. The models are partnered with WMT26 for the Video Subtitle Translation Task and the General Machine Translation Task.
The release provides high‑quality, efficient multilingual translation models with on‑device support via AngelSlim quantization.
Key changes
- Hy‑MT2 1.8B, 7B, 30B‑A3B models released
- Supports 33 languages and translation instructions
- AngelSlim 1.25‑bit quantization reduces 1.8B to 440 MB
- 1.5× speed improvement
- Outperforms DeepSeek‑V4‑Pro and Kimi K2.6
- Surpasses Microsoft and Doubao APIs
- IFMTBench benchmark released
- Translator Skill available on ClawHub and SkillHub
- Partnership with WMT26 Video Subtitle Translation Task