Zyphra has released ZONOS3-Mini, an open-source text-to-speech model designed for real-time use on local devices. The model features a sparse Mixture of Experts (MoE) architecture with 800 million total parameters and 100 million active parameters per token.
Key points
- Zyphra released ZONOS3-Mini, an open-source TTS model.
- The model has 100M active parameters and supports voice cloning.
- ZONOS3-Mini runs in real-time on local devices and browsers.
- It achieved 80ms median time to first audio on an H100 GPU.
ZONOS3-Mini supports zero-shot voice cloning and word-level delivery control, with weights available under Apache 2.0. Its compact size allows it to run in a browser tab, on a laptop, or on a server GPU.
The model can operate on various hardware, including NVIDIA GPUs via SGLang-Omni, Apple Silicon via MLX, and CPUs/GPUs via ggml. It can also run on phones or laptops using WebGPU in a browser.
By the numbers
- active parameters
- 100M
- total parameters
- 800M
- median time to first audio
- 80 ms
Zyphra reports that ZONOS3-Mini ranks first among open models on its ZTTS2-Eval Capability Index. The model achieved a median time to first audio of 80 milliseconds on an H100 GPU with 8 concurrent requests.
In browser-based performance, ZONOS3-Mini showed a median time to first audio under 570 milliseconds on a Pixel 9 and under 150 milliseconds on an M3 MacBook Air.
Zyphra trained ZONOS3-Mini on 112 NVIDIA H100 GPUs over 2.5 days. The model incorporates a new codec, Zyphra Audio Codec 1 (ZAC-1), which reportedly offers a 10x word error rate reduction compared to a DAC-trained model.
Sources and references
Original report
Zyphra Releases ZONOS3-Mini: Open-Source TTS With 100M Active Parameters
MarkTechPost
Read on MarkTechPost (opens in a new tab)Linked in the original report
- Linkzyphra.com/our-work/zonos3-mini (opens in a new tab)
- Modelhuggingface.co/Zyphra/ZONOS3-Mini (opens in a new tab)
- Linkzyphra.com/our-work/zonos2 (opens in a new tab)
- Linkzyphra.github.io/ZTTS2-Eval (opens in a new tab)
- Modelhuggingface.co/fishaudio/s2-pro (opens in a new tab)
- Modelhuggingface.co/OpenMOSS-Team/MOSS-TTS-v1.5 (opens in a new tab)
Ref 00Z4R5FU
Cite this
APA
Interestana AI Editorial. (2026, November 10). Zyphra Releases Open-Source TTS Model ZONOS3-Mini. Interestana. https://interestana.com/articles/zyphra-releases-zonos3-mini-open-source-tts-with-100m-active-parameters-00z4r5fu
MLA
“Zyphra Releases Open-Source TTS Model ZONOS3-Mini.” Interestana, 10 Nov. 2026, interestana.com/articles/zyphra-releases-zonos3-mini-open-source-tts-with-100m-active-parameters-00z4r5fu.
Chicago
Interestana AI Editorial. “Zyphra Releases Open-Source TTS Model ZONOS3-Mini.” Interestana, November 10, 2026. https://interestana.com/articles/zyphra-releases-zonos3-mini-open-source-tts-with-100m-active-parameters-00z4r5fu.
BibTeX
@misc{interestana_00z4r5fu,
author = {{Interestana AI Editorial}},
title = {{Zyphra Releases Open-Source TTS Model ZONOS3-Mini}},
howpublished = {Interestana},
year = {2026},
month = nov,
url = {https://interestana.com/articles/zyphra-releases-zonos3-mini-open-source-tts-with-100m-active-parameters-00z4r5fu},
note = {Summary of reporting by MarkTechPost, https://www.marktechpost.com/2026/11/10/zyphra-zonos3-mini-open-source-tts/}
}Credit the source
Original reporting: MarkTechPost, “Zyphra Releases ZONOS3-Mini: Open-Source TTS With 100M Active Parameters”, https://www.marktechpost.com/2026/11/10/zyphra-zonos3-mini-open-source-tts/(opens in a new tab)
Follow this story
Get Computing & Chips in your inbox — new stories in 40 words or fewer, once a week.
Subscribe