TMTPOST — Tencent Hunyuan released its next-generation speech recognition model Hy ASR 3.0 preview on Tuesday and made the engine available via Tencent Cloud APIs. The model draws on the language-understanding capabilities of the latest Hy3 large language model to produce more accurate, coherent transcriptions that better capture user intent amid complex real-world audio inputs. It has already gone live first on the Yuanbao product, where it is offered free of charge, while WorkBuddy and additional Tencent services are in the process of integration. Target applications include intelligent customer service, content understanding and voice search. The addition of the Hy-ASR-3.0-preview engine to Tencent Cloud’s real-time speech recognition WebSocket interface expands the company’s ASR portfolio with a system tuned for noisy or mixed-language conditions typical of production environments. By linking recognition performance directly to the Hy3 foundation model, Tencent positions the service as a practical upgrade for developers and internal products that require higher fidelity transcription without separate post-processing steps.