ZONOS3-Mini: Zyphra's open voice-cloning TTS that runs in a browser
An 800M-parameter sparse mixture-of-experts text-to-speech model with only ~100M active parameters per step, doing zero-shot voice cloning in 43 languages and running in a browser via WebGPU, on Apple Silicon, or on a server.
Why it matters
Shrinks Zyphra's previous 8B-parameter Zonos 2 to 800M total params while adding word-level control over emotion, pace and delivery, all under Apache 2.0 with no per-character bill.
What you could build with it
An indie developer building a language-learning app could give every lesson a consistent native-speaker voice by cloning one recorded speaker, generating all dialogue locally in the browser with no per-character API bill.
Does it hold up?
Usable today for local voice-cloning experiments: the browser WebGPU demo, MLX and ggml runtimes, and Apache 2.0 weights all work now, but independent quality evidence is thin and the headline benchmarks are Zyphra's own.
Built with ZONOS3-Mini
- ZONOS3-Mini Review: Voice Cloning in Your Browserarticle · Hands-on review: a Mac mini generated 38.5 seconds of audio in 2.81 seconds locally, with notes on the English-only demo and hardware dependence.
- Zyphra's ZONOS3-Mini Clones Voices With 100M Active Parameters, Runs in a Browserarticle · Release overview covering the 800M/100M-active MoE design, 43 languages, 27 stock voices, and the sub-150 ms time-to-first-audio figure on an M3 MacBook Air.
- Zonos 3 Mini: Free Open-Source Voice Cloning, Explainedarticle · Plain-language explainer on the free Apache 2.0 weights, browser WebGPU runtime, and the honest caveat that vendor benchmarks dominate the quality claims so far.
- sgl-project/sglanggithub · 38,894 ★ · SGLang-Omni lists day-zero serving support for ZONOS3-Mini, giving teams a production inference path for the model alongside its browser and MLX runtimes.
Learn more
First spotted on hf: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.