• News
  • Asia

Tencent Releases Hy4 Flash Weights for Local Deployment

Three checkpoints available for local deployment under Tencent Hy License

11/19/2026
•Ali Abounasr El Alaoui
Back to News

Tencent released the weights for its Hy4 flash AI model on Wednesday. The release came two days after the model became available through the company's apps and cloud API. The standard model has about 224 billion parameters, with around 12 billion active per token.


Downloadable versions

Tencent published three variants on Hugging Face. The set includes the standard instruction-tuned model, an FP8 quantized version, and a base model for further training. The standard checkpoint holds roughly 440.5 GB of tensor data, while the FP8 version holds roughly 287.1 GB.

License terms

The checkpoints use the Tencent Hy License 1.0. Model-as-a-service operators above US$50 million in aggregate worldwide revenue over 12 months need separate written permission. The same condition applies to products or services with more than 100 million monthly active users.

What the model does

Hy4 flash accepts text and image inputs and supports a context window of up to 1 million tokens. Tencent designed it for everyday productivity tasks such as research, document preparation, and spreadsheet work. Deployment materials include vLLM and SGLang container images, with examples using eight-way tensor parallelism.


Tencent says the larger Hy4 model's weight release is still being prepared. The current checkpoints are available for local deployment under the Tencent Hy License. Developers can use the published containers and guides to start working with the model.