Skip to content
View original post on X: MJ Kayum
Published
62/100

Tencent releases Hy4 preview under Apache 2.0 with FP8 and GGUF variants

AITencent's Hy4 preview is now available under the Apache 2.0 license, with official weights, an official FP8 version, and a mixed GGUF version. The model has 770B total parameters, 49B active, and native 1M context.
MJ KayumVerified on X
@mjkayum
View on X

FOR AI BUILDERS WHO PREFER OWNING THE STACK:

Tencent Hy4 preview is now available under Apache 2.0.

You can explore:

Official weights
https://huggingface.co/tencent/Hy4-preview

Official FP8
https://huggingface.co/tencent/Hy4-preview-FP8

Mixed GGUF
https://huggingface.co/AngelSlim/Hy4-preview-GGUF

The model is 770B total / 49B active with native 1M context.

Its mixed quant reduces the footprint from ~1.5TB to ~214GiB, while Tencent reports that
practical scores remain close to BF16.

Save the links if you build with open models.

And if you test it, share the real numbers—not just screenshots.

@TencentHunyuan @TencentAI_News

Source details

MJ Kayum · x.comPublished · added here

Original title: FOR AI BUILDERS WHO PREFER OWNING THE STACK:

Topics

Open-source ecosystemDeployment & engineeringChina AIModel releases

Tags

#Hugging Face