ElevenLabs introduced two new speech models on Monday, v4 and v4 Turbo, built on a new architecture that the company says improves expression control and speeds up voice cloning. The models support more than 90 languages, up from 70 in v3, with the largest quality gains in Japanese, Brazilian Portuguese, Mandarin, and Cantonese. Users can now stack multiple inline expression tags, and the model can start generating audio while an underlying LLM is still producing answers, which lowers latency for voice-agent use.
The launch comes as ElevenLabs reports strong commercial momentum. The company says more than 55% of its business now comes from large enterprises, and its annualized revenue run rate has roughly doubled from about $330 million at the start of the year to over $600 million. Headcount has grown past 800 across markets including India, Europe, and Brazil.
ElevenLabs raised $500 million earlier this year at an $11 billion valuation, and reports suggest a follow-up round could value it at $22 billion. In a recent interview, co-founder and CEO Mati Staniszewski said the company is targeting an IPO "in the next years," without committing to a specific timeline. The speech-model space is competitive, with startups like Cartesia, Deepgram, and WellSaid Labs, as well as Google and OpenAI, all pushing forward on expressive voice technology.