Tencent released the weights of Hy4, its flagship large language model, on November 20, making the 770-billion-parameter mixture-of-experts model available for download and deployment on GitHub, Hugging Face Inc., modelscope, and GitCode1,2. The release carries an Apache 2.0 license, permitting commercial use and modification.
The move completes the open-sourcing of the entire Hunyuan LLM series: Hy3-preview, Hy3, Hy4 preview, Hy4, and Hy4 flash are now all publicly available.
Hy4's mixture-of-experts language backbone contains 770 billion total parameters, with 49 billion activated per token. The model supports a context window of approximately one million tokens and handles text, image, and video understanding. Tencent says Hy4 improves on Hy4 preview in instruction following, long-context understanding, extended tasks, and research.
The model supports workflows spanning graphical interfaces and command-line tools, alongside frontend development, game development, and office tasks. Tencent provides deployment instructions for vLLM and SGLang, fine-tuning resources, and a safety evaluation card.
Hy4 initially launched on November 4; the November 20 weight release follows the separate opening of the smaller Hy4 flash model.
ANALYSIS By shipping the full Hy4 weights under Apache 2.0 alongside the lighter Hy4 flash, Tencent-Hunyuan now offers a dual-tier open-source stack: a cost-effective line for constrained deployments and a 770B flagship for high-end workloads, positioning the series as a direct alternative to other permissively licensed open-weight families.