Tencent’s Hunyuan team has released Hy Realtime, a new real-time voice model that understands speech, follows spoken instructions, talks in a human-like voice, calls tools to get things done, and can even sing. The model is already live in Tencent’s Yuanbao assistant and in Tencent customer service, with a developer API on Tencent Cloud coming soon.
Tencent announced the model on November 13 through its official Hunyuan account, as reported by Sina Technology. The release adds two headline capabilities to Tencent’s voice stack: agentic tool calling and singing, on top of stronger speech understanding, instruction following and human-like expression.
What Was Announced?
Hy Realtime is built for live conversation rather than one-shot questions. According to Tencent, it listens as you speak, understands what you mean even when sentences are vague, half-finished or corrected mid-sentence, and decides on its own when to search the web or call a tool before answering. The result is meant to feel less like a question-and-answer box and more like an agent you can talk to.
The singing capability is the most eye-catching addition. Tencent says the model can sing, moving its voice assistant beyond the usual question-and-answer pattern and toward richer, more entertaining interaction.
What Hy Realtime can do. Graphic: GPTOnline, based on Tencent Hunyuan’s announcement.
Where to Try It
Hy Realtime is already working inside Tencent products. It powers voice calls in the Yuanbao app, where users can try it by opening the latest version of the app, starting a voice call and choosing the “Chuyu” voice. It is also live in Tencent customer service, handling real support conversations. Tencent has also published a web demo where visitors can try some of the model’s abilities.
Developers will not have to wait long for access. Tencent says Hy Realtime will soon arrive on Tencent Cloud as an API service for developers and business customers.
Where Hy Realtime is available now, and what comes next. Graphic: GPTOnline.
What Comes Next
Tencent says Hy Realtime is still being improved, with three priorities for the next stage: duplex interaction, so the model handles the natural back-and-forth of live conversation more smoothly; long-term personalized memory, so it can remember a user’s habits and preferences; and interaction between foreground and background agents, extending what it can do on a user’s behalf.
Why It Matters
Real-time voice has become one of the most competitive fronts in AI, with OpenAI, Google and others all pushing voice agents that talk, listen and act. Tencent’s answer leans on its own distribution: Yuanbao gives Hy Realtime a large consumer audience at once, and the customer service deployment is a live test of whether a voice model can handle real users, not just demos.
The coming Tencent Cloud API is the part developers should watch. If Hy Realtime’s tool calling and singing hold up in third-party apps, Tencent will have a voice offering that goes beyond the usual chatbot, and one that arrives through the same cloud platform many Chinese businesses already use. The release also keeps up Hunyuan’s fast pace: it follows the November 4 launch of Hy4, Tencent’s largest open-weight model to date.
