According to CoinMeta, Tencent's Hunyuan has open-sourced a simultaneous interpretation model hy-simulmt that supports real-time bidirectional translation between Chinese and English with a delay of about 3 seconds while listening and translating. Additionally, the model can clone the speaker's voice tone. By integrating an additional TTS model, the system can broadcast in another language as well. The official team provided a deployment scheme based on qwen3-tts. In tests with long audio files, a streaming translation delay of around 3 seconds was measured, and the translation quality surpassed that of all the open-source models tested, as well as that of 4 out of 5 commercial API models. The team also enlisted 6 certified professional translators to conduct on-site tests with 18 segments of real speeches and interviews; the model's translation quality reached the level of professional simultaneous interpreters, with typically even lower delays. The model has a total of about 30 billion parameters, and approximately 3 billion parameters are activated with each inference. The weights, inference, and training tools have all been made available. Tencent's HY translation APP is also available.