The internal testing interface for Tencent’s cutting-edge large-scale model, Hy4, has been unveiled, signaling its imminent official launch. Tencent persists in a dual-track strategy of concurrent in-house model development and integration of open-source models, presenting three primary model choices: Hy4, Hy3, and DeepSeek. Hy4, specifically designed as an expert-level model, outstrips its forerunner, Hy3, in performance metrics. Its capabilities are anticipated to be further amplified through a trillion-parameter architecture or via reinforced reinforcement learning (RL) post-training techniques. Hy4 is slated to underpin the intelligent agent application, WorkBuddy, as its foundational technology moving forward. The preceding Hy3 model, boasting a 295-billion parameter framework, has already showcased exceptional prowess across diverse, demanding tasks, with its blind test results outperforming those of comparable models in the market.