Voice developers get free open models that clone a voice from three seconds of audio, which is useful and open to misuse.
Kenning's take
What it does
Qwen3-TTS is a family of open speech models from the Qwen team, released in January 2026 in 0.6B and 1.7B sizes. Base models clone a voice from about 3 seconds of audio, CustomVoice models offer nine preset voices with style control, and a VoiceDesign model creates voices from descriptions. They cover 10 languages including English, Chinese, German, French and Spanish, and are also offered through Alibaba's API.
- Voice cloning from 3 seconds of audio
- Nine preset CustomVoice timbres
- Voice design from text descriptions
- 10 languages plus dialect voices
- 0.6B and 1.7B sizes, Apache 2.0
Who it suits. Developers building voice apps who want an open, self-hostable speech model.
Key facts
- Category
- AI models
- Part of
- Qwen
- Made by
- Alibaba (Qwen team)
- Based in
- China
- Released
- 22 January 2026
- How to get it
- Free weights on Hugging Face and ModelScope; online demo; DashScope API
- Works on
- self-hosted, API, Hugging Face
- In the UK
- Yes
- Free plan
- Yes, open weights
FAQ
When was Qwen3-TTS released?
On 22 January 2026.
Can Qwen3-TTS clone a voice?
Yes. The Base models clone a voice from about 3 seconds of audio.
Which languages does Qwen3-TTS speak?
Ten, including English, Chinese, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
MAI-UI
Alibaba Tongyi research models that operate smartphone apps by looking at the screen and tapping
Qwen-Image-2.0
Alibaba's 7-billion-parameter model that both creates and edits images, with strong text rendering at 2K
Qwen-Image-Layered
Open Alibaba model that splits an image into separate transparent layers so each part can be edited alone
Qwen-Image-3.0
Alibaba's hosted image model built for text-heavy layouts, taking prompts up to 4,500 tokens