The collaborative team from Tencent Hunyuan and WorkBuddy has unveiled the 'Hy4 Preview Model Upgrade Announcement.' This upgrade comes in response to user feedback concerning prolonged thinking times and excessive self-verification during complex task execution. Through targeted improvements, these issues have now been comprehensively addressed and fully implemented. Following the optimization, the model achieves a notable reduction in both task rounds and token consumption, all while preserving task effectiveness. The Hy4 preview was launched and made open-source on August 28, boasting a total of 770 billion parameters and 49 billion active parameters, with a context length surpassing 1 million. It is primarily tailored for Agent, Coding, and productivity applications. On September 1, the team introduced a lightweight variant of the model, featuring compressed weights of roughly 214GB, thereby lowering the threshold for local deployment. According to OpenRouter's weekly data, the Hy4 preview secured the top spot globally, with 14.7 trillion tokens invoked and a month-on-month growth rate of 379%.