ElevenLabs Expands Voice AI Capabilities with v4 Model Launch

The v4 generation introduces refined control over voice identity, particularly during extended passages of text. By expanding the use of inline tags, the company allows users to layer multiple vocal expressions sequentially, giving the model a nuanced grasp of tone. These improvements extend to language support, with the startup noting significant quality gains in Mandarin, Cantonese, Japanese, and Brazilian Portuguese.

For enterprise clients, who now account for over 55% of the company’s business, the focus remains on latency. The v4 model integrates directly with LLM outputs, beginning audio generation the moment text tokens appear. This capability is designed to handle complex conversational dynamics, including escalations and customer service holds, with greater fluidity. With a valuation reaching $11 billion following a $500 million round led by Sequoia, and an annualized revenue run rate topping $600 million, the company is positioning itself for a future IPO as it faces intensifying competition from players like Cartesia, Deepgram, and OpenAI.