KyutaiInteractivity alignment
Moshika RL Seamless
Moshika RL Seamless is Kyutai's trained adaptation of Kyutai Moshika. Its weights are gated, and access requires accepting the release conditions.
Restricted access
The training targets when to speak
The study uses reinforcement learning to improve pauses, turn-taking, short acknowledgements and responses to interruption. A separate content reward helps retain useful answers. These are dialogue-behavior changes, not proof of better performance on every knowledge task.Post-training speech models for better interactivity
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable |
| Model identifiers | kyutai/moshika-rl-seamless |
| Terms | Public repository metadata specifies CC BY-NC 4.0.moshika-rl-seamless: Kyutai model card |
Related models
- Moshi Moshika
Kyutai ยท Moshi
A Moshi speech-dialogue model that listens while speaking, using Kyutai's Moshika voice.
Launch date not confirmedResearch releaseDownloadable