Qwen3-ASR-Flash, Qwen3-TTS-Flash and Qwen3-LiveTranslate-Flash (APIs)
September 2025 Qwen speech services on the API: 11-language ASR, TTS with 97 ms first packet, and 18-language live translation with visual context.
ASR-Flash (2025-09-08) builds on Qwen3-Omni with context biasing and singing recognition; TTS-Flash (2025-09-22) offers 17 voices across 10 languages and many Chinese dialects; LiveTranslate-Flash (2025-09-23) adds lip, gesture and on-screen-text cues and latency near three seconds.
- Date
- Monday, 8 September 2025
- Lab
- Alibaba (Qwen)
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| TTS-Flash first-packet latency | as low as 97 ms Single-threaded; Qwen claims best WER on seed-tts-eval and the MiniMax multilingual set vs MiniMax, ElevenLabs and GPT-4o-Audio-Preview | company |
| LiveTranslate-Flash latency / languages | about 3 s / 18 languages Plus dialect and accent translation for six Chinese varieties (company) | company |
Dates from Qwen's blog feed, in Beijing time, are 2025-09-08, 2025-09-22 and 2025-09-23. Benchmark leads are Alibaba's own comparisons. Open-weight Qwen3-TTS and Qwen3-ASR followed in January 2026.
Sources
This record was partly confirmed: some claims could not be checked on 6 October 2026. How we check