StepAudio 3 Realtime
StepAudio 3 Realtime combines audio understanding, turn control, reasoning and tool use. It supports simultaneous speech and can distinguish an interruption from a brief acknowledgement.
Preview
Continue a conversation during tool work
The model can begin a backend task after it understands the intent. The conversation can continue while the task runs, and the result can join the dialogue when it returns.StepFun stepaudio-3-realtime guide
Use Chat when the answer should be text
StepAudio 3 Chat offers the Audio 3 family's sound understanding through Chat Completions. It accepts audio or text by turn and returns text, rather than the duplex speech stream of Realtime.StepFun stepaudio-3-realtime guide
Practical guidance. Chat and Realtime are documented access paths. They are grouped here because the reviewed guides do not establish a separate trained base for each interface.
Separate interruption from silence
Practical guidance. A voice application needs to handle talk-over, short pauses and backchannel words. Evaluate these cases in a complete conversation. A fast first audio response alone does not show whether turn control works well.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | API, Provider product |
| Model identifiers | stepaudio-3-realtime-previewstepaudio-3-realtimestepaudio-3-chat-preview |
| Current access | The guide uses stepaudio-3-realtime-preview. The billing catalog announces the paid stepaudio-3-realtime name for 9 November 2026.StepFun billing catalog and 9 November 2026 audio change |
| Interaction | WebSocket Realtime API with managed conversation history and voice activity detection.StepFun stepaudio-3-realtime guide |
Related models
- Step-Audio R1.5
StepFun · Step-Audio R
Step-Audio R1.5 adds human-feedback training for natural dialogue while reasoning and speech proceed together.
Launch date not confirmedAPI · Provider product - StepAudio 2.5 Realtime
StepFun · StepAudio 2.5
StepAudio 2.5 Realtime accepts and returns speech over WebSocket.
Launch date not confirmedAPI · Provider product