The real-time version of Qwen3-Omni-Flash, built on the Thinker-Talker MoE architecture. It understands text, images, audio, and video, generates speech, and supports text interaction in 119 languages and speech interaction in 20 languages. This version is a snapshot from September 15, 2025.
Inference Service Provider
The inference service provider for qwen3-omni-flash-realtime-2025-09-15 is QwenCloud.
Model Capabilities
- Singapore
- China North 2 (Beijing)
| Capability | Support | Capability | Support |
|---|---|---|---|
Input Modality | Text Image Audio Video | Output Modality | Text Audio |
Model Experience | Unsupported | Function Calling | Unsupported |
Structured Outputs | Unsupported | Web Search | Unsupported |
Prefix Completion | Unsupported | Context Caching | Unsupported |
Batch Inference | Unsupported | Fine-tuning | Unsupported |
Context Limits
- Singapore
- China North 2 (Beijing)
| Parameter | Value | Parameter | Value |
|---|---|---|---|
Max Input Length | 56000 | Max Output Length | 8000 |
Context Window | 64000 |
Pricing
This page only shows the original pricing for model API calls, excluding any limited-time promotions. Visit QwenCloud console for promotional offers.
- Singapore
- China North 2 (Beijing)
| Billing Item | Price (USD) | Unit |
|---|---|---|
Input: Text | 0.52 | Per 1M tokens |
Input: Audio | 4.57 | Per 1M tokens |
Input: Image/Video | 0.94 | Per 1M tokens |
Output: Text (input contains only text) | 1.99 | Per 1M tokens |
Output: Text (input contains images/audio/video) | 3.67 | Per 1M tokens |
Output: Text & Audio (output text is not charged) | 18.13 | Per 1M tokens |
Rate Limits
- Singapore
- China North 2 (Beijing)
| Parameter | Value |
|---|---|
RPM (Requests Per Minute) | 60 |
TPM (Tokens Per Minute) | 100,000 |