Forbes reports that Tencent's newly released Hy3 model is a deliberate bet on AI agents rather than raw scale. Built on a mixture-of-experts design with 295 billion total parameters but only about 21 billion active, Hy3 is tuned for autonomous tasks like coding and enterprise workflows, and Tencent says it reaches a 90% task-completion rate across several internal apps. Caixin adds that demand since the launch has been strong enough that Tencent is expanding its AI computing capacity. The model is already wired into Tencent products including the Yuanbao assistant and CodeBuddy. It's another sign Chinese labs are chasing practical agent performance over benchmark supremacy.