On August 28, Tencent Hunyuan rolled out and open-sourced the preview version of its new-generation large language model, Hy4. This model is notable for its colossal scale, encompassing a total of 770 billion parameters, with 49 billion actively engaged in processing. It boasts a context length surpassing 1 million, showcasing outstanding performance in real-world productivity scenarios such as coding, office tasks, and scientific research applications. The Hy4 preview version has undergone a significant expansion in terms of model size, context length, and data scale, leading to a substantial enhancement in its intelligence capabilities. This positions it firmly within the top tier of open-source models available today.

Concurrently, the model has been open-sourced and integrated into Tencent's suite of products, including WorkBuddy/CodeBuddy, Yuanbao, and ima, making its debut across these platforms. Users now have the opportunity to directly interact with the model or leverage its API through Tencent Cloud Tokenhub and OpenRouter. Regarding pricing, the input cost is set at 6 RMB per million tokens, while the output cost stands at 18 RMB per million tokens. Additionally, cache hits are priced at 0.3 RMB per million tokens, offering a cost-effective solution for users.