ElevenLabs announced Eleven v4, a text-to-speech model, alongside Eleven v4 Turbo, a low-latency variant intended for use cases including voice agents. The company says v4 is designed to interpret tone, pacing, emotion, character, and context, and supports more than 90 languages. It also describes improved multi-speaker dialogue and more consistent voice identity.
ElevenLabs reports approximately 150 ms median time to first speech for v4 Turbo and approximately 100 ms median inference latency. These are company-reported figures; the announcement does not establish independent validation of the performance claims.