Tencent releases open-weight Hy4 preview

Tencent released Hy4 preview, a new-generation mixture-of-experts model aimed at software engineering, office analysis, game development and scientific research. The model card lists 770 billion total parameters, 49 billion activated per token, 78 layers and a one-million-token context. Full and FP8 weights are available through Hugging Face and other repositories, with vLLM and SGLang deployment guidance.

Why it matters: Sparse activation reduces the computation used per token relative to a dense 770B model, while the open weights broaden deployment choice. The footprint is still substantial, and claimed benchmark advantages have not been independently reproduced.

Tencent announcement | Official model card | a report