01
What is Qwen-Audio-3.1?
Qwen-Audio-3.1 is Alibaba Cloud Model Studio's current hosted audio family for speech recognition, live duplex voice interaction and generated audio. The important implementation detail is that these are separate models and endpoints, not one universal API.
The currently documented family includes qwen-audio-3.1-tts-next for generated speech and soundscapes, qwen-audio-3.1-realtime-plus for live voice applications, and Qwen-Audio-3.1 ASR variants for streaming and file transcription. Model availability and endpoint details vary by region.
02
TTS-Next and Realtime solve different jobs
TTS-Next is an AudioGen-style model that can generate speech together with dialogue, ambience and sound effects. Alibaba documents Chinese and English support, optional reference audio and completed audio outputs rather than a low-latency streaming assistant workflow.
Realtime Plus is designed for full-duplex spoken applications. It accepts audio and text and returns audio and text, supports turn detection and cloned voices, and can use either function calling or built-in web search; current documentation says those two capabilities cannot be enabled together.
03
ASR, pricing and regional limits
For transcription, Alibaba documents qwen-audio-3.1-asr-flash-streaming for live WebSocket recognition and qwen-audio-3.1-asr-flash-filetrans for asynchronous long-file transcription. File transcription supports recordings up to 12 hours or 2 GB, with speaker-diarization caveats for very long audio.
There is no single Qwen-Audio-3.1 price. For example, Alibaba currently lists TTS-Next Beijing pricing at $0.848 per million input tokens and $1.696 per million output tokens, while Realtime Plus uses separate text/audio input/output prices that differ by region. Always budget from the exact model and region page.
04
Access, open-weight status and limitations
The 3.1 audio models verified for this guide are accessed through Alibaba Cloud Model Studio. No open-weight release for this specific 3.1 audio family was verified in the first-party sources, so it should not be conflated with other downloadable Qwen model families.
TTS-Next is non-streaming and has narrower documented language coverage than ASR. Realtime and ASR have their own protocol and history limits. Launch percentage price-cut claims should remain attributed rather than being used as a substitute for the live model-specific price tables.
05
Which Qwen-Audio-3.1 model should you use?
Choose ASR when the output is a transcript, Realtime Plus when a user needs a live spoken conversation, and TTS-Next when the output is a produced audio asset that may combine voices, ambience and effects.
For production, pin the exact model ID, deployment region and protocol, then recheck current pricing and limits. Treat ASR-Next launch references separately from the callable ASR-Flash model IDs unless Alibaba publishes a matching first-party API page.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.