Overview
Tencent Hunyuan’s open-weight model for real-time, bidirectional Chinese–English simultaneous speech translation. It receives streaming audio and produces translated text while the speaker is talking. The 30B-A3B checkpoint provides translation-only output; a separate Interleave variant also outputs source text. (github.com)
Capabilities and uses
- Intended for live communication such as meetings, speeches and interviews.
- The supplied SimulInfer service supports a web interface, audio-file uploads and real-time WebSocket use. Speech synthesis and voice cloning require a separately configured TTS service. SimulTrain tools are also provided. (github.com)
- Tencent reports approximately 3 seconds of StreamLAAL on the RealSI long-form benchmark. This is a vendor-reported translation-lag metric, not time to first spoken output or a guarantee for every deployment. (huggingface.co)
Limitations and access
The stated simultaneous-translation directions are Chinese to English and English to Chinese. Translation quality and latency can vary with the audio and deployment. Human review remains important for high-stakes or accountability-sensitive use. The weights are published under Apache-2.0; check the terms of any separately used components. (huggingface.co)
Release date
October 26, 2026, according to Tencent’s release note. (github.com)
Pricing
As of November 15, 2026, model-specific official API input and output prices could not be verified in Tencent Cloud’s mainland China or international TokenHub price lists. No CNY or USD amount, billing unit, or hosted API availability is assumed. Self-hosting and any separate TTS service may incur costs. (cloud.tencent.com)