What Kyutai does
Kyutai is a Paris-based open-science AI lab focused on multimodal, real-time speech and audio intelligence, and on releasing models, code, and research for public use. Its flagship product direction is “voice AI for LLMs”: Kyutai built low-latency streaming speech-to-text (STT) and streaming text-to-speech (TTS) systems, then combined them into Unmute to let “any text LLM hear and speak” in real time by wrapping it with Kyutai’s STT/TTS components and using text prompts to control behavior. In parallel, Kyutai has released end-to-end speech and speech translation models such as Moshi and Hibiki, plus smaller-footprint voice models (e.g., Pocket TTS) intended for practical deployment. Beyond speech, Kyutai’s research portfolio also includes world-model work such as MIRA, with open-sourced datasets and training/inference code.
Kyutai’s business model is not presented as a traditional product-sales company because it is structured as a non-profit laboratory; instead, it is funded by philanthropic/strategic donors and publishes open research artifacts (model weights, inference code, technical reports, and demos) intended to support the broader AI ecosystem. Strategically, Kyutai positions itself as a contributor to open multimodal foundation model progress—especially in interactive, streaming audio settings where low latency and real-time behavior are central to user experience.