On October 26, Tencent's Hunyuan team released and open-sourced Hy-SimulMT, a simultaneous interpretation model supporting real-time speech-to-speech translation between Chinese and English with voice cloning, AI Daily reported. It debuted free on the Tencent Hy translation app with a measured response latency of about 3 seconds — faster than the 4–5 seconds of human professional interpreters.
On the RealSI long-audio benchmark, Hy-SimulMT posted the lowest latency and best open-source quality, outperforming most commercial APIs. Against six CATTI-certified simultaneous interpreters, it scored higher on translation quality and lower on latency across ten domains. The model will serve the APEC meeting in Shenzhen in November.
By squeezing response time into the human-expert range, the open-source model is rapidly lowering the barrier to real-time cross-language communication — a capability with clear diplomatic and business value.
The release adds to a wave of Chinese open-source models targeting practical, deployment-ready AI rather than benchmark showpieces.
Source: AI Daily / Tencent
