Tencent releases Hy-SimulMT for Chinese-English simultaneous translation
Original title: tencent/Hy-SimulMT-30B-A3B-FP8 — new model trending #29 on Hugging Face
The FP8 30B model streams speech both ways with voice cloning; Tencent reports about 3 seconds of StreamLAAL.
Tencent Hunyuan released Hy-SimulMT-30B-A3B-FP8 on October 26, 2026, for bidirectional Chinese-English speech-to-speech translation while the speaker is still talking. It produces translated text and speech and can clone the speaker's voice when a TTS service is configured. On the vendor's RealSI long-form comparison, it reported the lowest StreamLAAL shown and a GEMBA score above the listed open-source models and four of five commercial APIs. SimulInfer, SimulTrain and the Hy Translate App are available.
Unconfirmed: Latency and quality figures are Tencent's own comparisons, not an independent audit.
- Weights are published as tencent/Hy-SimulMT-30B-A3B-FP8 and were trending at number 29.
- Vendor table lists 90.3 on FLEURS, 75.5 on CoVoST2, 81.2 on ACL6060 and 84.7 on RealSI-Sentence.
- Speech output and voice cloning need a configured TTS service.
- StreamLAAL measures streaming lag, not time to first speech output.
- Vendors
- Tencent
- Products
- Hy-SimulMTHy Translate App
- Organizations
- Tencent Hunyuan
- AI models
- Hy-SimulMT-30B-A3B-FP8
From the original article
10 min read at huggingface.co<img src="assets/tencent-hy-logo. png" width="400" alt="Tencent Hy" /> <img src="assets/hy-simulmt-title. png" width="300" alt="Hy-SimulMT" /> <p align="center"><b>A Low-Latency, High-Quality Model for Real-Time Simultaneous Translation</b></p>
By tencent · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.