Eleven v3 (alpha) is a new Text to Speech model unveiled by ElevenLabs, offering unprecedented expressiveness and control in speech generation across 70+ languages. It introduces features such as multi-speaker dialogue, audio tags for controlling tone and emotion, and a new Text to Dialogue API for creating natural-sounding conversations. While it requires more prompt engineering compared to previous models, it significantly enhances expressiveness with capabilities like sighing, whispering, and laughing in speech. Designed for applications like videos and audiobooks, Eleven v3 is not recommended for real-time or conversational use cases due to its higher latency and need for optimization. A real-time version and better support for Professional Voice Clones are in development, with current availability through ElevenLabs' website and API, and an 80% discount offered until the end of June 2025 for self-serve users.