Tencent has launched Hy Realtime, its new real-time voice model, which excels in core conversational capabilities such as speech comprehension, instruction following, and human-like expression, while also introducing Agentic tool calling and singing features. The model is already available in products including Yuanbao voice calls and Tencent customer service, and will soon be accessible via Tencent Cloud API. It can leverage context to infer the intent behind incomplete input, maintain stable performance in multi-turn conversations, and autonomously decide whether to perform web searches. Its voice quality, emotional expression, and empathy have been comprehensively upgraded, and through collaboration with the Tencent Music team, it now supports song requests and singing in natural language. On the technical front, the model achieves industry-leading real-time interaction through optimizations including pre-training and post-training cross-modal alignment, a dedicated data system, multi-reward reinforcement learning, and low-latency speech generation. Future iterations will add capabilities such as duplex interaction and long-term personalized memory.