Kyutai (Moshi / Unmute)
Paris non-profit lab releasing open speech models: Moshi, Unmute and Pocket TTS
Overview
Kyutai is a non-profit open-science AI lab in Paris funded by Iliad, CMA CGM and Schmidt Sciences. It releases open speech models, including the full-duplex Moshi, the Unmute voice wrapper for text LLMs, the CPU-friendly Pocket TTS and Hibiki-Zero translation, with MIT or Apache code and CC-BY 4.0 weights.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Kyutai (Moshi / Unmute)
- Moshi full-duplex speech dialogue model with open weights
- Unmute, a speech wrapper that gives any text LLM a low-latency voice
- Kyutai speech-to-text and text-to-speech models
- Pocket TTS, a 100M-parameter model that runs on a CPU in seven languages
- Hibiki-Zero speech translation into English
- Free web demos of Unmute and Moshi
Kyutai is one of the most useful sources of open voice AI: Moshi for full-duplex conversation, Unmute for giving any text model a voice, and Pocket TTS for fast speech on a CPU, all with permissive code and CC-BY weights.
These are research releases, not supported products. Moshi's answers lag its latency, Unmute needs a 16 GB GPU to self-host, and commercial voice services now live at its spin-off, Gradium.
What we found
Kyutai is a non-profit open-science AI lab in Paris, launched in November 2023 with nearly €300 million committed at the start, including €100 million each from Xavier Niel's Iliad group and the shipping group CMA CGM; Schmidt Sciences is the third funder. It is led by Patrick Pérez, with Laurent Mazaré heading engineering and Alexandre Défossez research, and it gets its compute from Iliad's Scaleway. Everything it builds is released openly: code under MIT or Apache licences and model weights under CC-BY 4.0.
Its best-known work is in voice. Moshi, unveiled in July 2024 with weights released that September, is a full-duplex speech model that listens and talks at the same time, which Kyutai describes as the first speech-native dialogue system. Unmute, open-sourced in July 2025, takes a different route: it wraps any text LLM with Kyutai's own speech-to-text and text-to-speech models, so you can give an existing model a low-latency voice; self-hosting it needs a CUDA GPU with at least 16 GB. Pocket TTS, a 100-million-parameter model, runs about six times faster than real time on two CPU cores of a MacBook Air and now speaks seven languages. A community-trained Indonesian voice is listed in its repository. Hibiki-Zero translates speech from French, Spanish, Portuguese and German into English.
The lab has widened its scope in 2026: MIRA, a world model, in July; MuScriptor for music transcription; and KE:SAI, a physical-AI lab created with the ELLIS Institute Tübingen. Its first spin-off, Gradium, sells a commercial voice API separately.
Reception of Moshi's open release on Hacker News was largely critical of its answer quality while praising its latency. That fits the lab's purpose: it publishes research-grade systems for others to build on, not finished products. There is nothing to buy from Kyutai itself; the web demos at unmute.sh and moshi-chat.kyutai.org are free.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Kyutai (Moshi / Unmute)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Kyutai (Moshi / Unmute) unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Kyutai (Moshi / Unmute) directly →
Spotted something out of date? Suggest an update →
Alternatives to Kyutai (Moshi / Unmute)
More in Video & Audio
