Tencent officially announced the release of Tencent Hy4, its latest-generation large language model designed to compete directly at the pinnacle of China’s rapidly crowding AI landscape. The Hy4 preview is structured as a 770-billion-parameter Mixture-of-Experts (MoE) model that activates 49 billion parameters per token, balancing heavy computational muscle with deployment efficiency. Breaking from purely closed ecosystems, Tencent has made Hy4 available under the permissive Apache License 2.0, adding heavy fuel to the ongoing open-weight arms race against domestic rivals like DeepSeek, Alibaba, and Z.ai. The architecture features deep upgrades in visual understanding, complex chart analysis, mathematical reasoning, and multi-step tool orchestration tailored for enterprise and developer workflows. Tencent is already weaving Hy4 into its massive consumer and enterprise product suite ranging from workplace tools like WeCom and Tencent WorkBuddy to AI-driven NPCs and automated testing inside flagship games like Honor of Kings.

Full report : Tencent its massive 770-billion-parameter Mixture-of-Experts model under an Apache 2.0 license, targeting advanced multimodal reasoning, agentic workflows, and game production.