Tencent Hunyuan Open Source Hy4 Flash, three versions released at once
Tencent officially open-sourced Hy4 flash, just two days after the model release. This time, three versions of the model weights are released at once: Hy4 flash is the officially fine-tuned model that can be directly used for dialogue, reasoning, and agent tasks; Hy4 flash-FP8 is the 8-bit quantized version, mainly used to reduce memory usage during deployment; Hy4 flash Base is the pre-trained base model without instruction fine-tuning, making it convenient for developers to continue training and customization.
Hy4 flash has a total of 224 billion parameters, with 12 billion activated during each inference. Even after FP8 quantization, the deployment threshold remains high. Tencent has made it available for download on GitHub, Hugging Face, ModelScope, and GitCode, and the community has also created a GGUF quantized version for easier operation in different hardware environments. The model uses the Tencent Hy License, allowing developers to use and modify the weights according to the agreement, but with licensing conditions; large model API service providers must obtain separate authorization if they reach the specified revenue or user scale thresholds. As for the flagship Hy4 with 770 billion parameters, Tencent stated that it is still accelerating the open-source process and has not announced a specific timeline.