Real-time multimodal
This section provides API references for Qwen-Omni-Realtime. qwen3.8-omni-flash-realtime, qwen3.5-omni-plus-realtime, and qwen3.5-omni-flash-realtime support WebSocket, WebRTC, and AOQ. For historical model qwen3-omni-flash-realtime-2025-09-15, see Python SDK and Client Events for its WebSocket documentation. The applicable scope of media modalities and event fields is determined by their respective access and event references.
Protocol selection
The following comparison describes the general characteristics of the protocols. The data types in the table indicate what each protocol can transmit; the specific inputs and outputs supported by each model are subject to the access documentation. Access difficulty, performance over weak networks, and connection speed are qualitative references; actual experience depends on network conditions, devices, and implementation.
| dimension | WebSocket access | WebRTC access | AOQ access |
|---|---|---|---|
| Use cases | Server-side integration, rapid prototyping validation | Browser-side interaction, traditional audio/video calls | AI multimodal real-time interaction, weak-network scenarios, hybrid data transmission |
| Browser compatibility | Native support | Native support | Not supported |
| Integration effort | Very low | Moderate | Low |
| Weak-network resilience | Poor | Good | Excellent |
| Data transmitted by the protocol | Text, audio, image | Audio/video, text | Audio/video, text |
| Connection speed | Slow | Slow | Fast |
| Echo cancellation and noise reduction | Handled by the client | Built-in | Built-in |
| AI scenario adaptation | Basic, suitable for pure text or scenarios with less stringent real-time requirements | Traditional design; AI scenarios require additional adaptation | Natively tailored for AI multimodal data characteristics |
| Client-side platforms | Environments supporting WebSocket are supported | Browsers, mobile devices | Android, iOS, HarmonyOS; other SDK versions are listed on the access page |
Browsers natively support WebSocket, but authentication requirements must still be met when accessing the model. Do not store long-term API keys in browsers or on mobile devices.
AOQ (AI over QUIC) Client SDK is available for Android, iOS, HarmonyOS, Windows, macOS, Electron, and Linux. For downloads and platform-specific API, see AOQ Access.
Events and other references
Voice configuration
All three protocols can be used for Token Plan. Available models and regions are subject to the specifications outlined in Token Plan Personal Edition and Team Edition.