Zyphra · On-Device TTS

ZONOS3-Mini

Zyphra has released ZONOS3-Mini, an open-weight zero-shot voice-cloning text-to-speech model built for real-time on-device use. It is a sparse mixture-of-experts model with 800M total and about 100M active parameters, and it runs via WebGPU, MLX, ggml and SGLang. Zyphra reports 80 ms median time to first audio on one H100 at 8 concurrent requests and a Capability Index score of 66.2 on its own ZTTS2-Eval benchmark, ahead of Fish Audio S2 Pro (61.7). The weights are under Apache 2.0, and all benchmarks were run by Zyphra and not independently verified.