Tencent Hy (open source models)

Tencent Hy is Tencent’s open-weight foundation model series, published under Tencent’s Hugging Face organization (tencent/...) for self-hosted execution and community research. The open lineup encompasses large language models (LLM), vision-language models (VLM), and visual generation models (VGM).

Model Series & Checkpoints

• Hy4 Series: The flagship generation, introducing native vision input alongside LLM reasoning, long-context understanding, and deep-research capabilities. Model weights are published under open licensing with ecosystem integrations for vLLM and SGLang. Derived scientific foundation models include HyCell (single-cell representation) and HyPDE (physical field modeling).
• Hy3 & Hy3 Preview (tencent/hy3): Open-weight general language checkpoints supporting a 256,000-token context window and 128,000 max output tokens.
• Hy-MT2 Series (tencent/hy-mt2-1.8b, -7b, -30b-a3b): Specialized, parameter-efficient translation models configured with 8,192-token context windows.
• Visual Generation Lineup: Includes HY-World 2.0 and Hy-Nomas for multi-modal image, video, and world-model generation tasks.

Because model weights and code repositories are openly distributed, organizations can deploy Hy checkpoints on private GPU infrastructure using standard serving frameworks. (For Tencent's hosted API console and enterprise application layer, see the separate Tencent Hy Studio listing.)