A model for real-time audio and video interaction with text and audio output, multichannel audio, and tool calling.
Inference service provider
The inference service provider for qwen3.8-omni-flash-realtime is Alibaba Cloud Model Studio.
Model capabilities
Supported regions: China (Beijing) and Singapore. Use an API key from the selected region.
| Capability | Support | Capability | Support |
|---|---|---|---|
| Input modalities | Text, streaming audio, and video (consecutive image frames) | Output modalities | Text and audio |
| Function Calling | Supports custom tool calling | MCP | Supports remote MCP tools. See the MCP interaction flow |
| Multichannel audio | WebSocket access supports 1, 2, or 4 channels; multichannel input uses 16 kHz PCM | Video aggregation | Configure session.video.input.representation_compact: the default is none; set it to normal to enable aggregation |
| Voices | The default is Tina. See voice parameters and audio previews |
Access the model through WebSocket, WebRTC, or AOQ. See the real-time guide for examples and SDK configuration.
For WebSocket access, use a workspace-specific WebSocket endpoint. For configuration, see Establish a connection.
Context limits
| Parameter | Value | Parameter | Value |
|---|---|---|---|
| Maximum total input tokens | 196608 | Audio history | Up to 100 turns and 600 seconds in total |
| Video history | Up to 50 turns and 240 seconds in total | Maximum output tokens | 65536 |
Older history is discarded when the turn count or cumulative media duration exceeds a limit. Cumulative media duration is not the total session duration.
Model pricing
See Model pricing for inference prices.
For speech output, qwen3.8-omni-flash-realtime bills both the audio and its corresponding text at their respective output rates. Qwen3.5-Omni-Realtime models bill only the audio; the corresponding text is not charged.
Free quota: 1 million tokens, available only in Singapore. Valid for 90 days from service activation, model release, or application approval, whichever is later. See Free quota.
China (Beijing)
| Billing item | Price (USD) | Unit |
|---|---|---|
Input: audio | 0.848 | Per million tokens |
Output: audio | 1.696 | Per million tokens |
Input: text/images/video | 0.212 | Per million tokens |
Output: text | 0.636 | Per million tokens |
Singapore
| Billing item | Price (USD) | Unit |
|---|---|---|
Input: audio | 0.93 | Per million tokens |
Output: audio | 1.87 | Per million tokens |
Input: text/images/video | 0.23 | Per million tokens |
Output: text | 0.7 | Per million tokens |
Rate limits
Both Beijing and Singapore have limits of 60 RPM and 2,000,000 TPM. See Rate limits for model request limits.