Tencent has open-sourced three versions of Hy4 flash; the flagship Hy4 has not yet announced its open-source release date.
BIBIBI
AT A GLANCE
Tencent Hunyuan has open-sourced three versions of Hy4 flash; the flagship Hy4 with 770B parameters has not yet announced a specific open-source release date.
Article
Two days after Tencent released Hy4 flash, it open-sourced the models, releasing all three versions at once; the flagship Hy4 will have to wait.
Beating AI Flash News: Tencent Huan Yuan has officially open-sourced Hy4 flash, just two days after the model's release. This time, the weights for three versions were released simultaneously:
• Hy4 flash: The official model fine-tuned with instructions, which can be directly used for dialogue, reasoning, and agent tasks.
• Hy4 flash-FP8: An 8-bit quantized version, primarily used to reduce GPU memory usage during deployment.
• Hy4 flash Base: A pre-trained base model without instruction fine-tuning, making it easy for developers to continue training and customization.
Hy4 flash has a total of 224B parameters, with 12B activated per inference. Even after FP8 quantization, the deployment threshold remains high. Tencent has made the models available for download on GitHub, Hugging Face, ModelScope, and GitCode, and the community has also created GGUF quantized versions to facilitate running on various hardware environments.
The model uses the Tencent Hy License, which allows developers to use and modify the weights according to the agreement, but comes with licensing conditions. Large-scale model API service providers must obtain separate authorization if they meet specified revenue or user scale thresholds.
As for the flagship Hy4 with 770B parameters, Tencent stated that it is still accelerating the open-source process and has not yet announced a specific timeline.
The report said that Tencent Hunyuan open-sourced Hy4 flash two days after its release and released model weights for three versions at once.
02
Hy4 flash is a model fine-tuned on instructions, suitable for dialogue, reasoning, and agent tasks.
03
Hy4 flash-FP8 is the 8-bit quantized version, mainly used to reduce GPU memory usage during deployment.
04
Hy4 flash Base is a pre-trained foundation model that has not been fine-tuned with instructions, allowing developers to continue training and customization.
05
The report says that Hy4 flash has a total of 224B parameters, with 12B activated per inference.
06
The report says the model can be downloaded from GitHub, Hugging Face, ModelScope, and GitCode; the community has also created a GGUF quantized version.
07
The report says that the model uses Tencent Hy License, and the use or modification of weights under the agreement is subject to licensing conditions; large model API service providers that reach the specified revenue or user scale thresholds must obtain separate authorization.
08
The report said that Tencent stated it is still accelerating the open source of its flagship Hy4 with 770B parameters, but did not announce a specific timeline.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
The report introduces three versions of the Hy4 flash: the instruction-tuning version that can be used directly for tasks, the FP8 quantization version aimed at reducing memory usage, and the Base version that facilitates further training. It also mentions the model size, download channels, and licensing conditions. The open-source release date for the flagship Hy4 remains unknown.
Why it matters to readers
Open weights provide developers with options to download and continue training and customization; however, reports indicate that deployment still has barriers, and licenses may also require some large API providers to obtain separate authorization.
If you are a developer, you can first differentiate the uses of the three versions and check the licensing conditions. Reports indicate that even when using the FP8 quantization version, the deployment barrier remains high.
Risks and unknowns
The report does not provide a direct verification link to Tencent’s official announcement.
The specific content of the licensing conditions, the prescribed threshold values, and the scope of application were not elaborated in the materials.
The specific open-source release date for the flagship Hy4 has not yet been announced.
There is no documentation on the developers' actual adoption and deployment costs.
Related Developments
Loading event timeline…
Related concepts
Instruction Fine-tuning
This term is not in the glossary yet. Browse related concepts in the glossary.