Tencent's Hunyuan team announced on November 13 the release of Hy Realtime, a real-time speech model built for actual phone-style conversation rather than clean, read-aloud prompts. The company says the model can reconstruct intent from the messy parts of human speech — mumbled words, half-finished sentences, mid-stream self-corrections — and that its multi-turn instruction following and context management have been strengthened so it keeps track of what a user meant across long, winding conversations.
The agentic part is what separates it from a typical voice assistant. Based on the conversation and the user's intent, the model itself decides whether a web search is needed, then handles the search decision, the tool call and the spoken answer as one continuous process. Tencent frames this as unifying what used to be three separate steps — deciding to search, calling the tool, generating the answer — into a single flow inside the voice call itself.
On the output side, Tencent reports improvements over its predecessor in audio quality, pronunciation, prosody and emotional expression, with more natural shifts in tone over the course of a conversation. It says Hy Realtime shows significant advantages in internal expert A/B tests on search-and-answer tasks, an internal reasoning test set for non-thinking mode, and leading results on several public voice benchmarks — all company evaluations; the published materials do not give sample sizes or test settings, so those figures should be read as vendor-reported and have not been independently verified.
The headline feature, inevitably, is singing. Developed together with an AI team at Tencent Music Entertainment (TME), the model lets users request songs by artist, album, genre, language, release year or popularity, combine those conditions, and refine the request in follow-up turns. In Tencent's example, a user asks for an English folk song released between 2010 and 2020, and the model picks a match and sings it. Tencent says it tuned the singing for expressiveness, natural phrasing and breath, and for melody and rhythm.
Availability is already live in two places: the latest Yuanbao app, where users can choose the voice named Chuyu in voice calls, and Tencent's customer service line. A Tencent Cloud API is coming soon, and a web demo is available on Tencent Cloud for developers who want to try the model's capabilities before the API lands. Tencent says the next steps for the line are duplex interaction, long-term personalized memory, and coordination between a foreground conversational agent and background agents.
The release fits the pattern of Chinese consumer AI assistants racing to make voice the primary interface. Where Western voice modes mostly answer questions, Hy Realtime is being positioned as an assistant that can act — search, call tools, carry out multi-step requests — and even perform, all without leaving the call. Whether the singing is a gimmick or a hook, the underlying bet is the same one animating the whole industry: the assistant people actually keep talking to will be the one that feels less like a menu and more like a person who can get things done.
Comments (0)
Log in to join the discussion
Log InNo comments yet