ElevenLabs' new v4 speech model makes AI voices more expressive and consistent
decoder ·
ElevenLabs has released its new Eleven v4 speech model, which significantly improves voice consistency and expressiveness across long productions. The updated architecture allows for more accurate interpretation of direction cues, enabling reliable generation of emotions, laughter, and sound effects. It supports over 90 languages and enhances voice cloning capabilities, including native accent retention in foreign languages. A new Turbo variant targets real-time applications like voice agents, starting speech production in just 150 milliseconds. Benchmarks indicate that v4 outperforms competitors such as Cartesia Sonic 3.6 and Google's Gemini 3.8 Flash TTS in both speed and pronunciation accuracy. The company is offering temporary price reductions on API usage through October 12 to encourage adoption. Both standard and Turbo models are now available via the API, ElevenAgents, and ElevenCreative platforms.
decoder