Alibaba's Tongyi Lab Launches "Qwen-Audio-3.0-ASR" Speech Recognition Models — Covers 7 Chinese Dialect Groups and 30 Languages

On July 30, 2026, Alibaba’s Tongyi Lab released a new automatic speech recognition (ASR) lineup called “Qwen-Audio-3.0-ASR.” It follows the “Qwen-Audio-3.0-TTS” text-to-speech model launched earlier the same month (July 20), rounding out the Qwen-Audio 3.0 generation with a recognition counterpart. The lineup ships three variants — a real-time streaming model, a file-transcription model, and a standard model — covering China’s seven major dialect groups plus 20+ regional accents, and 30 languages overall.

Details

• Three variants: qwen-audio-3.0-asr-flash-streaming for real-time streaming recognition, qwen-audio-3.0-asr-flash-filetrans for asynchronous transcription of long audio files, and a general-purpose qwen-audio-3.0-asr-flash
• Chinese dialect coverage: Recognizes seven major dialect groups — Mandarin, Wu, Xiang, Gan, Hakka, Min, and Yue (Cantonese) — plus 20+ regional accents
• Language coverage: Supports 30 languages including Chinese, English, Japanese, and Korean
• Specialized recognition features: Optimized accuracy for classical Chinese poetry, plus enhanced inverse text normalization (ITN) for numbers and dates
• Free quota: Some reports indicate a 10-hour free usage allowance
• Availability: Offered as a hosted API via Alibaba Cloud Model Studio (DashScope) and the Qianwen AI platform

How to try it

• Accessible via the DashScope SDK through Alibaba Cloud Model Studio
• No open-weight release has been announced; usage currently requires going through the hosted API