Tencent Open-Sources Hy4 with 770B Parameters and Million-Token Context
Tencent's Hunyuan team has officially open-sourced the Hy4 model weights, a 770B-parameter MoE architecture with 49B activated parameters and a context window exceeding one million tokens, alongside the cost-efficient Hy4 flash. This completes the full open-sourcing of the Hunyuan LLM series, with Hy4 and Hy4 preview released under Apache 2.0 for commercial use, plus a system card detailing its defensive-only cybersecurity training approach.
Related Entities
Full Report
IT之家 November 20 news: The Tencent Hunyuan team announced today that the Hy4 large model is officially open-sourced. The model weights have been simultaneously released on four platforms—GitHub, Hugging Face, ModelScope, and GitCode—and developers can download and deploy them on their own.
With this, the Hunyuan LLM series has been fully open-sourced, covering versions including Hy3 preview, Hy3, Hy4 preview, Hy4, and Hy4 flash.

The official Hy4 version was first released on November 4, and what is open-sourced this time is the model weights that were previously announced. Hy4 adopts a MoE (Mixture of Experts) architecture, with 770B total parameters, 49B activated parameters, and a context window exceeding 1 million tokens.
As IT之家 previously reported, alongside the flagship Hy4, the cost-effective Hy4 flash was also released. The Hy4 flash model has 224B total parameters, including 64B Engram parameters, activates 12B parameters per token, supports multimodal input and a 1 million token context length, and its API price is significantly lower than that of Hy4.
Regarding the open-source license, Hy4 and Hy4 preview use the Apache 2.0 license, permitting commercial use and modification. Tencent Hunyuan also released the Hy4 model System Card, disclosing evaluation results in dimensions such as cybersecurity and biosecurity.
According to the official explanation, Hy4's training data only includes defensive cybersecurity content such as PoC generation and vulnerability remediation, and excludes offensive data such as exploit development, reverse engineering, and penetration testing; audits by Tencent Xuanwu and Keen Security Lab found no obvious offensive capability risks.
References
- [GitHub] https://github.com/Tencent-Hunyuan/Hy4
- [Hugging Face] https://huggingface.co/tencent/Hy4
- [ModelScope] https://modelscope.cn/models/Tencent-Hunyuan/Hy4
- [GitCode] https://ai.gitcode.com/tencent_hunyuan/Hy4
- [GGUF] https://huggingface.co/AngelSlim/Hy4-GGUF
- [Open-source license] https://huggingface.co/tencent/Hy4/blob/main/LICENSE
- [Model System Card] https://hy.tencent.ai/Hy4_system_card.pdf