Qwen3-TTS Full Family Open-Sourced with Voice Design and Cloning

January 22, 2026

Details

  • Qwen has released the complete Qwen3-TTS family as open-source models, including VoiceDesign, CustomVoice, and Base variants.
  • The release includes five models across two parameter sizes: 0.6B and 1.8B, designed for different deployment scenarios and resource constraints.
  • Qwen3-TTS supports 10 languages (Chinese, English, German, Italian, Portuguese, Spanish, French, Japanese, Korean, and Vietnamese) and includes dialect variations.
  • The models feature free-form voice design and voice cloning capabilities, enabling users to create custom synthetic voices without pre-defined constraints.
  • Qwen3-TTS achieves state-of-the-art performance with a 12Hz tokenizer for streaming synthesis, delivering low-latency speech generation comparable to commercial services like ElevenLabs and MiniMax.
  • All models are available as open-source releases, eliminating licensing fees and enabling on-premises deployment and customization.

Impact

Qwen's full open-source release of Qwen3-TTS democratizes enterprise-grade text-to-speech technology by providing models that match or exceed commercial competitors in quality while eliminating per-minute API costs. The inclusion of voice design and cloning at open-source scale removes a significant technical and financial barrier that previously favored closed platforms like ElevenLabs, MiniMax, and Google's audio offerings. The dual model sizes (0.6B and 1.8B) enable deployment flexibility across edge devices and data centers, addressing both latency-sensitive and accuracy-focused use cases. This release reinforces Qwen's position as a serious contender in the foundation model space beyond language models, signaling Alibaba's commitment to building comprehensive AI infrastructure accessible to developers. The 10-language support and streaming architecture position Qwen3-TTS for rapid adoption in multilingual applications, potentially accelerating the shift away from commercial speech synthesis vendors. Over the next 12-24 months, this commoditization of TTS may redirect funding and developer attention toward differentiated applications rather than underlying synthesis technology, similar to how open-source LLMs reshaped language model economics.

Read more