Fact · Voice and AI
200milliseconds
Moshi, a full-duplex speech-to-speech model from Kyutai, has a theoretical latency of 160 ms and about 200 ms in practice.
Checked against its source on October 17, 2026
In the source's own words
“with a theoretical latency of 160ms, 200ms in practice”
Moshi: a speech-text foundation model for real-time dialogue, Alexandre Défossez and and colleagues at Kyutai, arXiv (Kyutai), 2024, Abstract.
Open the original- What it measures
- Moshi is a research model with a text language model as its backbone. The authors report latency for speech input and speech output in their own setup.
- What it does not show
- The figure is the authors measurement for the model. It does not include phone networks, tool calls or the checks a commercial agent adds.