ZONOS3-Mini: what it is and is it free? (Nov 2026)
Zyphra's ZONOS3-Mini is a free, open-source text-to-speech and voice cloning model that runs in your browser. Features, languages, limits.
Checked and updated 15 November 2026 at 04:00 UTC

Key points
- ZONOS3-Mini is a small open-source text-to-speech model from Zyphra that reads text aloud in 43 languages and can clone a voice from an audio sample.
- It's free to download under the Apache 2.0 license and runs entirely in a WebGPU browser, on a laptop or on a phone, with no account or server needed.
- Zyphra also hosts it on Zyphra Cloud for developers, but it hasn't published a price on the model's blog or Hugging Face pages.
ZONOS3-Mini is a free, open-source text-to-speech (TTS) model from Zyphra that turns text into natural speech in 43 languages and can clone a voice from a short recording. It's small enough to run in real time on phones and laptops, including right inside your web browser, and the weights are released under the Apache 2.0 license.
Zyphra announced it on November 9, 2026, and it's one of the most-liked new models on Hugging Face this week, with about 350 likes and nearly 20,000 downloads.
What ZONOS3-Mini does
According to the Hugging Face model card, you can:
- Read any text aloud in one of 27 built-in voices
- Clone a voice by uploading, recording or picking an audio sample
- Design a character voice by shaping age, accent and vocal texture
- Steer emotion, pace and delivery, down to single words, for games, narration and voice assistants
Zyphra's blog says emotions such as happy, angry or scared are targets the model aims for, not guarantees.
Is it free?
| How you use it | Cost | What you need |
|---|---|---|
| Zonos Voice Studio in your browser | Free, runs on your own GPU | A browser with WebGPU |
| Download the weights (Apache 2.0) | Free, commercial use allowed | A computer; GGUF files are about 660 MB |
| Zyphra Cloud (hosted API) | Not published on the pages we checked | A Zyphra Cloud account |
The browser version has two modes: the full Voice Studio and a Lite page for quick voice cloning. The first visit downloads the model, then your browser keeps it cached.
How small and fast is it?
ZONOS3-Mini is a "mixture of experts" model with 800 million parameters in total, of which only 100 million are active at once. That's why it fits on a phone. Zyphra reports under 150 milliseconds to the first audio on an M3 MacBook Air and under 570 milliseconds on a Pixel 9.
Developers can also run it on NVIDIA GPUs with SGLang, on Apple Silicon with MLX, or with a C++ build for CPUs and other GPUs.
How it compares
Zyphra says ZONOS3-Mini ranks first among open models on its own Zyphra TTS Capability Index, ahead of Fish S2 Pro and MOSS TTS v1.5, and it shows side-by-side samples against Fish Audio, Qwen, Cartesia and ElevenLabs. These are Zyphra's own tests, not independent ones.
What it means for you
If you want a free AI voice for a video, an audiobook draft, or a game character, ZONOS3-Mini lets you do it without a subscription and without your text leaving your computer. With nothing to install for the browser version, it's an easy open option to try before paying for a voice service. Use voice cloning only with your own voice or with clear permission.
For more on running AI locally, read on-device vs cloud AI and AI models you can run on your phone. Our free AI tools list has more no-cost options.
AI Tools This Week is not affiliated with Zyphra.
Frequently asked questions
Is ZONOS3-Mini free?
Yes, the model itself is. The weights are on Hugging Face under Apache 2.0, which allows commercial use, and the in-browser Zonos Voice Studio runs on your own GPU. Only the hosted Zyphra Cloud version may cost money, and Zyphra hasn't listed that price on the pages we checked.
Can ZONOS3-Mini clone my voice?
Yes. You upload, record or pick a short voice clip and it speaks new text in that voice. You can also design a new character voice by shaping age, accent and vocal texture. Only clone voices you have permission to use.
What do I need to run ZONOS3-Mini?
For the browser version, a browser with WebGPU support. The first visit downloads the model and the browser caches it. Zyphra reports under 150 ms to first audio on an M3 MacBook Air and under 570 ms on a Pixel 9.
Sources
Related
Haru-4B: the open AI voice model that talks and listens at once
Haru-4B from Japan's higurashi is a free, open voice model that listens and speaks at the same time, in 9 languages. What it does and how to run it.
Idol image-to-3D model: what it is and is it free?
Speridlabs' Idol turns one photo into a 3D Gaussian splat, including the sides you can't see. Free under Apache 2.0, but it needs an NVIDIA GPU.
Qwen3.8-27B: what it is, is it free and can you run it?
Alibaba's Qwen3.8-27B is a free, open AI model that runs on a strong home PC. What it does, file sizes, how to install it and benchmarks.