Step-Audio R1.1
Step-Audio R1.1 extends R1 with a dual-brain design. A formulation component handles higher-level reasoning while an articulation component handles speech generation.
Let thinking and speech overlap
The card links the design to Mind-Paced Speaking research. Its purpose is to allow reasoning to continue during spoken output rather than placing all reasoning before speech.StepFun Step-Audio-R1.1 model card
Check coherence across the spoken answer
Practical guidance. Overlapping reasoning and speech can reduce perceived waiting, but it also makes consistency important. Listen for whether later speech corrects or contradicts an early statement. Keep the downloaded checkpoint distinct from the newer hosted service.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Step-Audio-R1.1 |
| License | Apache 2.0 is declared in the model card.StepFun Step-Audio-R1.1 model card |
| Reasoning method | The card describes iterative self-distillation grounded in acoustic representations.StepFun Step-Audio-R1.1 model card |
| Hosted successor | The current API overview directs R1.1 users to the upgraded R1.5 service.StepFun audio model overview |
Related models
- Step-Audio R1
StepFun · Step-Audio R
An audio-reasoning model trained with Modality-Grounded Reasoning Distillation to retain acoustic evidence.
Launch date not confirmedRestricted accessDownloadable - 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 3 Realtime
StepFun · StepAudio 3
StepAudio 3 Realtime combines audio understanding, turn control, reasoning and tool use.
Launch date not confirmedPreviewAPI · Provider product