What's this about?
StepAudio 3 Realtime runs private reasoning alongside spoken output and asynchronous tools; authors report 56% τ-Voice task success and 98.9 on Full-Duplex Bench (94 HF upvotes).
What exactly is full-duplex voice?
A system can listen while speaking, so interruptions and backchannels feel natural. Fluency alone does not establish that its tool actions or answers are correct.
What's the community saying?
94 HF upvotes; paper metrics are author-reported.
- Significance: parallel reasoning can reduce conversational pauses.
- Significance: tool accuracy must be tested alongside voice flow.
- Significance: comparisons require matched latency/cost settings.