Tencent Hunyuan releases Hy-SimulMT translation model with about 3-second response

On 26 October 2026, Tencent Hunyuan released and open-sourced Hy-SimulMT, a real-time simultaneous interpretation model built on the Hy-SimulMT-30B-A3B specification. According to the Hugging Face listing, the model handles bidirectional speech-to-speech translation between Chinese and English. It continuously receives incoming audio and translates while the speaker is talking, producing translated text and speech at the same time. It also supports real-time voice cloning, preserving the speaker's vocal characteristics, and is available as GGUF releases for local deployment, under the names Hy-SimulMT-30B-A3B-GGUF and Hy-SimulMT-30B-A3B-Interleave-GGUF.

A second report, published the same day by IT Home, draws on the official introduction, which emphasizes low latency and high accuracy. Audio from the speaker can be fed into the model in real time; the model listens and translates as it goes, returning translated text and speech immediately, and the real-time voice cloning is said to reproduce the speaker's voice and tone faithfully. Latency is the headline claim: the measured response time is about 3 seconds, versus an average of 4-5 seconds for professional human interpreters. On RealSI, a widely used open benchmark for long audio, Hy-SimulMT posts the lowest latency by the StreamLAAL metric, while translation quality is scored by GPT6.1-Sol. According to the official claim, it reaches the best result among open models and surpasses most commercial simultaneous interpretation APIs. The model debuted in the Tencent Hy translation app, where it is offered free of charge.

The public record so far covers the launch and open-source release, with both reports citing Tencent's own figures; no independent third-party verification of the benchmark results has appeared yet.