Overview
Qwen-Audio-3.1-Realtime-Flash is a real-time duplex speech conversation model. It accepts text and audio and produces text and audio. It supports function calling, web search and voice cloning. The official QwenCloud model-release announcement is dated October 27, 2026. (docs.qwencloud.com)
Capabilities and uses
- Designed for real-time voice conversations, including voice assistants and customer-service applications.
- Supports WebSocket, WebRTC and AOQ access. Interaction modes include
server_vad,smart_turn,smart_turn_v2and push-to-talk; the documentation recommendssmart_turn_v2. - Web search is enabled with
enable_searchand can be used alongside function calling. The default voice islonganqian_v3.1. (docs.qwencloud.com)
Limits
- Context window: 262,144 tokens; maximum input: 245,760 tokens; maximum output: 16,384 tokens.
- Audio history in context: up to 50 turns and 300 seconds of cumulative audio.
- Thinking mode via
enable_thinkingis not supported. Speaker enhancement is supported insmart_turnandsmart_turn_v2modes. - The published Singapore and Beijing limits are 60 RPM and 100,000 TPM. For WebSocket, RPM counts connection requests; there is no separate concurrent-connection limit. (docs.qwencloud.com)
Pricing
The following are official model API prices per million tokens, checked on November 15, 2026. Input and output are billed separately. For text-only responses, the text-output rate applies. When a response contains both text and audio, only the audio output is charged; accompanying text has no separate output charge. (help.aliyun.com)
| Site and region | Currency | Text input | Audio input | Text output | Audio output |
|---|---|---|---|---|---|
| Alibaba Cloud domestic site, China (Beijing) | CNY | 1.5 | 6 | 4.5 | 12 |
| Alibaba Cloud domestic site, Singapore | CNY | 1.677 | 6.781 | 5.104 | 13.636 |
| QwenCloud international site, Singapore | USD | 0.23 | 0.93 | 0.70 | 1.87 |
| QwenCloud international site, China North 2 (Beijing) | USD | 0.212 | 0.848 | 0.636 | 1.696 |
These are separately published site- and region-specific prices, not currency conversions of one another. Official pricing sources: Alibaba Cloud domestic model page and QwenCloud model page. (help.aliyun.com)
Alibaba Cloud's international pricing page also lists a Singapore free quota of 1,000,000 tokens for this model, valid for 90 days from the latest of Model Studio activation, model release or application approval; it states that no free quota is available in other regions. Conversation history is billed as input again in subsequent turns. (alibabacloud.com)