From：Internet Info Agency 2026-08-31 12:52:00
On August 31, Tencent HunYuan announced that since the large language model Hy4preview was first integrated into the WorkBuddy platform on August 28, overwhelming user and developer enthusiasm has led to task queue backlogs on its launch day. To ensure a smooth user experience, Tencent HunYuan has urgently scaled up the Hy4preview inference cluster and will continue dynamically allocating resources. However, due to limited total high-end computing capacity and concentrated peak-time concurrency, occasional queuing may still occur during certain periods. Hy4preview has a total of 770 billion parameters, with 49 billion activated parameters, and supports a context length of up to 1 million tokens. The model is available for free until September 10. During this period, each user receives a daily free quota; once usage limits are reached, the interface will display the time remaining until the quota resets. Tencent recommends that users switch to other models such as Hy3 while waiting in queue—Hy3’s free access on WorkBuddy has been extended through September 30. Currently, Hy4preview does not support multimodal generation capabilities such as image or video creation. When users initiate multimodal tasks like image or video generation on WorkBuddy, the system automatically routes these requests to other multimodal models. These tasks consume credits under standard billing rules and do not count against the Hy4preview free quota.
