Zyphra released ZONOS3-Mini, an open-source text-to-speech model, on November 9, with weights on Hugging Face under Apache 2.0. It is a sparse MoE with 800 million total and about 100 million active parameters per token, trained on 112 H100s in 2.5 days. By Zyphra's numbers, time to first audio is under 570 ms in the browser on a Pixel 9 and under 150 ms on an M3 MacBook Air, and it ranks first among open models on the capability index of its new ZTTS2-Eval benchmark (35 languages of read speech plus 20 in-the-wild), ahead of Fish S2 Pro and MOSS TTS v1.5. It supports voice cloning, age, accent and gender steering, and word-level delivery control, with WebGPU, SGLang, vLLM, GGML and MLX runtimes.
Key Takeaways
- ✓Scale: 800M total, about 100M active parameters (64 experts, top-2 routing), versus 8B for the previous ZONOS2; open weights under Apache 2.0
- ✓Speed (Zyphra figures): under 570 ms time to first audio in a Pixel 9 browser and under 150 ms on an M3 MacBook Air; 80 ms median at 8 concurrent requests and peak 457 seconds of audio per second on one H100
- ✓The new ZAC-1 codec runs at 15 Hz versus 86 Hz for DAC and cuts word error rate about 10x under similar data; one generation can produce up to 10 minutes of 44.1 kHz audio
- ✓Evaluation: ZTTS2-Eval covers 35 languages of read speech plus 20 in-the-wild; on its 420-item capability index ZONOS3-Mini ranks first among open models, followed by Fish S2 Pro and MOSS TTS v1.5 (Zyphra's own benchmark, not yet independently reproduced)
- ✓Get started: try it in the browser via WebGPU with no install, or run
sgl-omni serve --model-path Zyphra/ZONOS3-Minifor an OpenAI-compatible /v1/audio/speech endpoint
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and the industry pain point
Voice agents, game dialogue and audio content usually force a choice between closed TTS APIs (per-character billing, data leaves your stack) and self-hosting large models (expensive memory, high latency). Zyphra's previous ZONOS2 traded 8B parameters for quality and was hard to run on device. ZONOS3-Mini aims to make open TTS run in real time on a phone.
Architecture and mechanism
The backbone has 11 hybrid layers, 9 Gated DeltaNet plus 2 global attention, with a 64-expert top-2 latent MoE. The new ZAC-1 codec drops the frame rate from 86 Hz to 15 Hz with 1 semantic and 9 acoustic codebooks. Speaker representation moves to MECT-B2, and DPO then GRPO post-training improves robustness. Word-level delivery control works by monitoring alignment attention heads, and a separate edit model can rewrite generated sentences.
Benchmarks and measured results
These are Zyphra's figures: under 570 ms first audio in a Pixel 9 browser and under 150 ms on an M3 MacBook Air; first among open models on its ZTTS2-Eval capability index, ahead of Fish S2 Pro and MOSS TTS v1.5. The benchmark is vendor-built and not yet independently reproduced, so test on your own languages.
Getting started for developers
The fastest path is the official WebGPU Voice Studio in the browser. On servers, SGLang-Omni starts an OpenAI-compatible endpoint in one command, and a references parameter clones a voice. Apple Silicon can use the MLX server and CPUs the GGML build. Text normalization covers 9 languages including English, Mandarin, Japanese and Korean.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.