- Who it helps
- Developers and creators evaluating local narration or voice interfaces
- Availability
- November 12 technical report; public code and Apache-2.0 model weights. The model card lists forty-three supported languages. Hardware and download requirements vary by runtime.
Active parameters are only part of the model
The report describes 886 million total parameters, with about 100 million active during each step. The smaller active figure does not describe the whole model download or memory footprint.
Choose a runtime before estimating speed
The repository describes CPU and GPU options, including a C++ runtime, Apple MLX and browser WebGPU. The browser uses a compatible GPU. Real-time performance claims were not checked on a modest CPU.
Evaluate the voice you intend to use
Use permissioned reference audio and check pronunciation, timing and speaker similarity on your own text. The report’s evaluation sets and the model card’s language list are different measures.
What you can do
- Read the primary description and its stated limits before planning an evaluation. Open reference ↗
Where the evidence stops
Documentation only; no model downloaded, speech generated or latency measured. No useful independent video was verified.
Sources read. Public source text checked November 14, 2026 UTC. No installation, model inference or complete workflow executed.
Sources & further reading
Hugging Face · official
ZONOS3-Mini model card ↗
Model card lists forty-three supported languages, Apache-2.0 weights and browser/native options. No audio generated.
Zyphra · page checked 2026-11-14
Paper · paper
ZONOS3-Mini technical report ↗
November 12 report describes 886 million total parameters and 100 million active parameters. Reported quality and speed remain author evidence.
Zyphra · page checked 2026-11-14 · Published 2026-11-12
More sources reviewed (1)
GitHub · implementation
ZONOS3-Mini runtime repository ↗
CPU, CUDA, Metal, Vulkan, MLX and browser runtime choices described. Device speed and setup commands were not tested.
Zyphra · page checked 2026-11-14
Coverage note: Documentation only; no model downloaded, speech generated or latency measured. No useful independent video was verified.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
ZONOS3-Mini generates speech and supports voice conditioning through browser and native runtimes.