ElevenLabs released Eleven v4 and its low-latency v4 Turbo variant on September 28, with the announcement updated October 1. The company describes an entirely new text-to-speech architecture, stronger emotional direction and support for more than 90 languages.
Availability and latency are separate claims
The release says the models are available through ElevenCreative, ElevenAgents and ElevenAPI. ElevenLabs reports roughly 100 milliseconds median inference latency and 150 milliseconds time to first speech for Turbo. Its comparison subtracts network latency, so this is not an end-to-end user latency guarantee.
The major audio generation is distinct from the company's employee tender. Expressiveness, cloning consistency and comparative speed remain provider claims; this review did not independently reproduce them.