What is Qwen3-TTS?
No dated launch filed yetApache-2.0
Summary
Apache-2.0 speech models from the Qwen team, 0.6B and 1.7B, with voice clone, voice design, and 10 languages.
- Category
- Voice models
- Open source
- Yes
- License
- Apache-2.0
- Models
- Qwen3-TTS-12Hz-1.7B, Qwen3-TTS-12Hz-0.6B
- Pricing
- Weights and code are Apache-2.0. No hosted price is quoted on the repository README we used.
- Links
- GitHub
Qwen3-TTS is the voice model this site would reach for first when the license has to be permissive and the feature list has to include cloning and design. The repository is QwenLM/Qwen3-TTS. The technical report is arXiv:2601.15621.
The README's feature list, which we are not extending:
- Release date in the news section: January 22, 2026, for the 0.6B and 1.7B models on the 12Hz tokenizer.
- Ten languages, named above.
- Base models clone a voice from about three seconds of audio.
- The 1.7B VoiceDesign model takes a natural-language description of the voice.
- Streaming: the README says end-to-end synthesis latency can be as low as 97 ms.
It is ranked in Best free AI voice models in 2026. There is no separate "launch" file because the directory's launch type is aimed at agent products; the model still has a hub so the guide can link "what is Qwen3-TTS?".
Qwen3-TTS alternatives
| Tool | License | Pricing note |
|---|---|---|
Qwen3-TTSThis pageApache-2.0 speech models from the Qwen team, 0.6B and 1.7B, with voice clone, voice design, and 10 languages. | Apache-2.0 | Weights and code are Apache-2.0. No hosted price is quoted on the repository README we used. |
VoxCPMApache-2.0 tokenizer-free TTS from OpenBMB. VoxCPM2 is a 2B model with 30 languages, voice design, cloning, and 48 kHz output. | Apache-2.0 | Weights and code are Apache-2.0, including commercial use, per the repository. |
ChatterboxMIT-licensed TTS from Resemble AI. Chatterbox Multilingual V3 is a 500M model for 23 or more languages with zero-shot cloning. | MIT | The model card lists an MIT license. No API price is copied here. |
KokoroApache-2.0 TTS model with 82 million parameters. Version 1.0, published January 27, 2025, lists 8 languages and 54 voices. | Apache-2.0 | Apache-2.0 weights. The model card says API hosts were under $1 per million characters as of April 2025. Self-hosting has no license fee. |
MOSS-TTSApache-2.0 speech family from OpenMOSS. MOSS-TTS-v1.5 supports 31 languages, with separate models for realtime streaming and voice design. | Apache-2.0 | The repository says models in the family are Apache-2.0. |
Launches and updates
No dated launch has been filed for Qwen3-TTS yet. The description above is from the project’s own pages.
Guides covering Qwen3-TTS
Questions
What is Qwen3-TTS?
Qwen3-TTS is a series of speech models released by the Qwen team on January 22, 2026. The README describes voice clone, voice design, and natural-language voice control. Sizes are 0.6B and 1.7B. The license is Apache-2.0.
Which languages does Qwen3-TTS support?
The README lists 10 languages: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian, plus dialectal voice profiles.
What are free alternatives?
VoxCPM2, Chatterbox, Kokoro, and MOSS-TTS are the other permissive voice models on this site. Kokoro is much smaller and does not clone a voice.