K

    Kyutai

    A French non-profit open-science laboratory dedicated to multimodal AI research and open-source foundation models.

    FRFrance·Founded 2023
    Headquarters
    🇫🇷France
    Data Hosting
    France
    Structure
    Independent
    Compliance
    ▣GDPR◆EU HQ▦EU Data

    Kyutai key facts

    Last updated
    Legal entity
    Kyutai
    Headquarters
    France (FR)
    Ownership
    Independent
    Founded
    2023
    Data hosting
    France
    Infrastructure
    Scaleway
    Deployment
    Self-hosted
    Open source
    No
    US Cloud Act exposure
    Low (no US parent)
    Replaces
    OpenAI Whisper, ElevenLabs, Google Cloud Speech-to-Text, Meta Llama

    About Kyutai

    Kyutai is a Paris-based non-profit laboratory dedicated to open research in artificial intelligence. Launched in November 2023 with €300 million in funding from the iliad Group, CMA CGM Group, and Schmidt Futures, it positions itself as Europe's first independent AI open-science lab. The organization focuses on developing and releasing open-source models that push the boundaries of multimodal AI, specifically in speech and language processing. The laboratory is known for its high-efficiency research and development, producing breakthrough models like Moshi—the first speech-native dialogue system—and Helium, a series of multilingual language models. By providing the model weights, codebases, and training methodologies freely to the community, Kyutai supports the European AI ecosystem and offers a transparent alternative to the 'black box' models of dominant US-based commercial entities. Research is conducted using the Nabu 23 superpod infrastructure provided by Scaleway, ensuring a sovereign European computing stack.

    Alternative to

    Target Segments

    Nonprofit / NGOEducationOther

    Deployment Options

    Self-hosted

    Hosting & Security

    Infrastructure

    Scaleway

    Key Features

    Moshi: Speech-native dialogue system for real-time voice interaction
    Helium 1: Multilingual 2-billion-parameter language model
    Hibiki: Real-time speech-to-speech translation system
    Pocket TTS: Lightweight text-to-speech with voice cloning
    Unmute: Low-latency streaming pipeline for LLM voice interaction
    ARC-Encoder: Context compression for large language models
    MoshiVis: Voice-based interaction about images

    Related Articles

    All articles