Tencent’s Voice Model Learned to Sing and to Call Tools

Hy Realtime decides on its own whether to search the web mid-conversation, and adds singing built with Tencent Music.

Tencent Hunyuan released the real-time speech model Hy Realtime on 13 November 2026, claiming gains in speech understanding, instruction following and humanlike expression, plus two new capabilities: agentic tool calling, in which the model judges from context whether to run a web search and unifies search decisions, tool calls and answer generation into one pass, and singing, co-designed with Tencent Music’s Lingyin AI team so users can request songs by artist, album, genre, language, release year or popularity and continue the thread across turns. It is already live in Yuanbao voice calls and Tencent customer service, with Tencent Cloud API access to follow. The commercial read is that voice agents are crossing from conversing to transacting, which sets up a direct comparison with per-minute priced rivals and puts pressure on outsourced customer service contracts.