Domestic GPU handles a 770-billion-parameter large model! Muxi Xiyun C-series Day 0 adaptation for Tencent Hy4 preview

Tencent Hunyuan officially released and open-sourced its next-generation large language model, Hy4 preview, today. Muxi's曦Cloud C-series GPU completed Day 0 adaptation for the model, achieving stable operation with high throughput and low latency through underlying operator optimization and software-hardware co-acceleration.

Hy4 preview has a total of 770B parameters (770 billion) with only 49B activated parameters (49 billion), and a context length exceeding 1M. The model adopts a MoE (Mixture of Experts) architecture, delivering outstanding performance on real productivity tasks such as coding, office work, and science, firmly placing it in the top tier of open-source models.

In a blind test involving 163 internal Tencent experts across 203 engineering tasks, Hy4 preview scored 2.99/4.00 on average, slightly outperforming GLM-5.3's 2.92 and Kimi K3's 2.94.

The Day 0 adaptation relies on Muxi's self-developed MXMACA full-stack software stack. This stack covers the entire chain, including underlying drivers, user-space interfaces, the MXCC compiler, deep operator adaptation, and mainstream framework integration, compressing traditional model adaptation cycles from weeks to hours.

The collaboration between Muxi and Tencent Hunyuan has established a stable rhythm of "release-ready adaptation." Since December 2025, Muxi has completed Day 0 adaptation for 34 leading mainstream models.

For the full Tencent Hunyuan model series, including Tencent-HY-MT1.5, Hunyuan Image 3.0, Hy-MT2, Hy3 preview, Hy3, and up to Hy4 preview, Muxi has achieved same-day adaptation on release.

Hy4 preview is now available simultaneously in Tencent products such as WorkBuddy, CodeBuddy, Yuanbao, and ima. Users can also access the API via Tencent Cloud Tokenhub and OpenRouter.

Pricing is set at 6 yuan per million input tokens and 18 yuan per million output tokens.

The completion of this adaptation further validates the maturity of Muxi's "self-developed GPU hardware + MXMACA full-stack software" integrated architecture in the deployment of domestic large models.