Gradium develops audio-native foundation models that unify speech generation, transcription, transformation and dialogue in a single neural architecture. It sells real-time multilingual text-to-speech, speech-to-text and voice cloning APIs to developers and enterprises, launching with English, French, German, Spanish and Portuguese. The company was founded in 2025 by Neil Zeghidour, Olivier Teboul, Laurent Mazare and Alexandre Defossez, who previously built the Kyutai research lab. Customers at launch spanned gaming, healthcare, customer support and market research.