MODELS
by Zyphra
Open-source MoE text-to-speech model for real-time, high-fidelity voice cloning with Apache 2.0 weights and hosted or self-hosted inference paths.
ZONOS2 is Zyphra’s open-source mixture-of-experts text-to-speech model for real-time, high-fidelity voice cloning. Official sources describe Apache 2.0 weights, a Hugging Face model page, a GitHub repository, and both hosted cloud and self-hosted inference paths. Zyphra Cloud is described as a full-stack platform with serverless inference and dedicated inference capacity for running open-weight models without managing infrastructure.
Use ZONOS2 when evaluating open-source text-to-speech models for real-time high-fidelity voice cloning. Use it when you need a model with both self-hosted options via the GitHub/Hugging Face materials and hosted deployment through Zyphra Cloud. Use it when Apache 2.0 model weights are important for your evaluation or deployment review.
Last updated Jun 14, 2026