Elevenlabs is releasing Eleven v4, a speech model supporting over 90 languages that improves consistency across long productions up to 10,000 characters. Its Turbo variant enables real-time voice agents to start speaking in approximately 150 milliseconds.