Tencent Hunyuan announced today that since Hunyuan Hy4 preview was first integrated into WorkBuddy on August 28, 2026, its significantly improved Agent capabilities have received enthusiastic feedback and active testing from users and developers. Queues appeared in the WorkBuddy task queue on the first day after launch, and Tencent apologized for this.
The announcement states that to ensure a better user experience, Tencent Hunyuan has urgently expanded the Hy4 preview inference cluster and will continue to dynamically allocate resources. However, due to the limited total amount of high-end computing power and concentrated peak concurrency, queues may still occur at certain times.
If a queue occurs, users are advised to consider the following temporary options:
Switch to Hy3 or other models: Hy3 is currently free on WorkBuddy, with the free-use period extended until September 30, 2026, 23:59.
Use the service during off-peak evening hours for non-urgent requests.
Two notes regarding free use of Hy4 preview:
Daily free quota is limited: Hy4 preview is free until September 10, 23:59. Due to limited resources, the platform assigns each user a daily free quota during this period. After the rate limit is triggered, the page will display the reset and recovery time.
Hy4 preview is a large language model and currently does not support multimodal generation such as images and videos. When you initiate image-generation, video, or other multimodal tasks in WorkBuddy, the system automatically switches to another multimodal model for execution. These tasks consume points according to the standard rules and do not use the free Hy4 preview quota.
Tencent officially released Hy4 preview on August 28. It has 770 B total parameters, 49 B activated parameters, and a 1M context length. According to the official statement, Hy4 preview has undergone significant expansion in model size, context length, and data scale. Joint advances in pretraining and post-training have brought another major improvement in intelligence, placing it firmly among the top tier of open-source models.