Tencent Hy is Tencent's open-weight model family. Hy4 flash launched on November 17, 2026 in instruct, FP8 and base variants, with weights published on Hugging Face, ModelScope and GitCode.
3 news · open weights · data as of 2026-11-22
Hy4 flash is a cost-oriented multimodal MoE model: about 220B total parameters (including 63.5B Engram), roughly 12B active per token, text and image input, and a 1M context window. Tencent positions it for low-cost everyday productivity.
Tencent's Hy team has open-sourced Hy4, Hy4-FP8 and the pretrained Hy4-base weights on Hugging Face. Hy4 is a new-generation multimodal Mixture-of-Experts flagship with a 770B-parameter language backbone, 49B text parameters plus 1.23B vision parameters active per token, text/image/video understanding and a 1,048,576-token context under the Tencent Hy License.
Tencent's Hy team has published Hy4-base on Hugging Face, the pretrained base version of its new-generation Mixture-of-Experts flagship model, supporting text, image and video understanding with a 770B-parameter language backbone. Weights are downloadable under the custom Tencent Hy License, with mirrors on ModelScope and GitCode and a GGUF variant.
Tencent Hunyuan published Hy4 flash, FP8 and base open weights: about 220B total parameters (including 63.5B Engram), roughly 12B active per token, text and image input, and a 1M context window under the Tencent Hy License.
The weights are publicly downloadable under the custom Tencent Hy License; open-weight is the more precise term. Commercial use and redistribution follow that license.
How large is Hy4 flash? +
About 220B total parameters including 63.5B Engram, roughly 12B active per token, with a 1M context window.
Where can I download Hy4 flash? +
Hugging Face, ModelScope and GitCode host the instruct, FP8 and base weights.