Tencent Open-Sources Hy4 Preview, a 770-Billion-Parameter Model for Coding, Research, and Enterprise Work

Tencent's Hunyuan team released and open-sourced Hy4 preview on August 28, a next-generation mixture-of-experts large language model with 770 billion total parameters (about 49 billion active per request) and a context window exceeding 1 million tokens. Tencent said Hy4 preview is the first Hunyuan model to have participated in optimizing its own training pipeline, reportedly lifting end-to-end throughput on Tencent's internal training and inference systems by 31.8%. Aimed at coding, office productivity, data analysis, game development, and scientific research, the model scored an average 2.99 out of 4 across 203 engineering tasks judged by 163 outside experts, which Tencent said outperformed comparable open models. It is available on Hugging Face and via Tencent Cloud's TokenHub and OpenRouter, with API pricing of $0.834 per million input tokens and $2.501 per million output tokens, and free two-week trials through Tencent's WorkBuddy and CodeBuddy products. The release adds Tencent to the list of major Chinese tech firms — alongside DeepSeek, Alibaba, and Zhipu AI — racing to ship large open-weight models as domestic compute and coding-assistant competition intensifies.