By Hamza Faisal
Tencent's Hunyuan team has released Hy Realtime, a real-time voice model that can hold a natural spoken conversation, decide on its own when a question needs a web search — and break into song when asked. The model went live on November 13, 2026, and it is already answering calls inside the company's Yuanbao assistant and its customer service lines.
The launch was announced through Hunyuan's official WeChat account and reported by Chinese outlets including Yicai, National Business Daily and IT Home. It is the latest step in Tencent's push to make its Hunyuan models the backbone of everyday products, following the release of its flagship Hy4 model earlier in November.
A voice model that decides when to search
Hy Realtime is built for live conversation rather than one-off questions. According to Tencent, it understands spoken language, follows instructions and speaks with human-like expression — and it adds a layer of judgement on top. When a caller asks something that needs fresh information, the model decides for itself whether to run a web search and does it in the middle of the conversation, instead of waiting to be told.
That agentic tool calling is the model's headline feature. Instead of simply answering from what it already knows, Hy Realtime can reach for live information when the question demands it, then fold the result back into the spoken reply. Tencent says the model also handles the basics of a good voice assistant: recovering the speaker's intent, keeping track of the conversation and responding in a natural tone.
It can sing, too
The model has one feature its rivals rarely advertise: it can sing. The singing capability was co-designed with Tencent Music Entertainment's TME Lingyin AI team, and Hy Realtime will break into song when a user asks it to. It is a deliberate showcase of the model's control over tone, rhythm and expression — the same qualities that make a spoken voice sound human rather than robotic.
Tencent says Hy Realtime scored leading results on several public voice benchmarks in its own internal blind tests, though it has not published sample sizes or test settings, so the claims remain the company's own until independent testing appears.
Where to try it — and what comes next
Hy Realtime is already live in voice calls in the Yuanbao app, where users can select the voice named Chuyu, and in Tencent's own customer service. A web demo is available on Tencent Cloud, and the company says an API on Tencent Cloud will follow, without giving a date or a price.
Tencent says the model is still being refined. Next steps include duplex interaction — where both sides can speak at once, as in a real conversation — longer-term personalised memory, and coordination between foreground and background agents working on a task together.
The launch lands in a crowded race for real-time voice, with OpenAI's voice mode and Google's Gemini Live already competing for the same users. Tencent's bet is that a voice assistant which searches, sings and takes real customer calls is more useful than one that only talks — and with Hy Realtime already inside its own products, it will not have to wait long to find out.
