ElevenLabs has released Eleven v4, a new speech model designed to improve emotional expression and consistency across long-form audio productions. The update introduces a redesigned architecture that enables the AI to follow direction cues more accurately, generate nuanced sounds like laughter or whispers, and support more than 90 languages.
The release includes the Eleven v4 Turbo variant, optimized for real-time voice agents in customer service and gaming applications, which delivers a response latency of 150 milliseconds. Eleven v4 handles up to 10,000 characters per request, improves voice cloning capabilities with native accents across languages, and features enhanced pronunciation controls that score 91.7 percent on benchmarks. Both models are available now through the company's platform and API with temporary promotional pricing.
This release matters because it narrows the trade-off between speed and emotional expression in real-time conversational AI while improving localization quality for global media productions.