Model reference · open weights

moshika-rl-seamless

moshika-rl-seamless is an open-weight audio or speech model from kyutai. moshika-rl-seamless (BF16) weighs 15.8 GB; the smallest configuration that runs it is RTX 4090 24 GB.

• Moshika-rl-seamless is a 7.7B parameter audio-to-audio model developed by kyutai for full-duplex spoken dialogue.
• It is post-trained with reinforcement learning to improve interactivity, specifically targeting pause handling, turn-taking, backchanneling, and user interruption.
• The model supports English and is licensed under CC BY-NC 4.0.

Summary of the kyutai/moshika-rl-seamless model card, 2026-10-01

Hugging Face ↗GitHub ↗Paper ↗