Eleven v3 is the latest generation of text-to-speech model launched by ElevenLabs. It is positioned as the most emotionally expressive speech synthesis system currently available. According to the official documentation, this model can generate highly natural and emotionally varying voices, and supports complex multi-speaker conversation scenarios. This model supports over 70 languages and introduces the "Audio Tags" mechanism, allowing users to directly control the emotion, tone and presentation style of the voice through text tags. At the same time, Eleven v3 supports multi-character dialogue generation, which can be used to build natural voice interactions or story-based audio content.
Scribe v1 is the first speech-to-text model launched by ElevenLabs. The official description of it is that it is "the most accurate transcription model in the world". This model was officially released on February 26, 2025. Its core design goal is to cope with the unpredictability of real-world audio, and it can handle conference recordings, phone calls, podcasts, and even audio in noisy environments.
This model is a dedicated model recommended by ElevenLabs for scenarios involving low-latency, conversational voice intelligent agents. Eleven Flash v2 is an ultra-low latency text-to-speech (TTS) model launched by ElevenLabs.
Eleven Flash v2.5 is a high-speed text-to-speech model provided by ElevenLabs, specifically designed for real-time speech generation scenarios. According to the official documentation, this model has been optimized in terms of speech generation speed and system response time, and is currently one of the TTS models with the lowest latency. Flash v2.5 supports approximately 32 languages, and can handle up to about 40,000 characters in a single request, providing an extremely low latency performance of approximately 75 milliseconds (excluding network transmission and application layer delays). Therefore, it is highly suitable for applications that require high real-time performance, such as voice assistants, conversational AI, game interaction systems, and large-scale automatic speech generation tasks.
Eleven Flash v2.5 is a high-speed text-to-speech model provided by ElevenLabs, specifically designed for real-time speech generation scenarios. According to the official documentation, this model has been optimized in terms of speech generation speed and system response time, and is currently one of the TTS models with the lowest latency. Flash v2.5 supports approximately 32 languages, and can handle up to about 40,000 characters in a single request, providing an extremely low latency performance of approximately 75 milliseconds (excluding network transmission and application layer delays). Therefore, it is highly suitable for applications that require high real-time performance, such as voice assistants, conversational AI, game interaction systems, and large-scale automatic speech generation tasks.
Eleven Turbo v2.5 is a text-to-speech model provided by ElevenLabs. The official description of it is that it is a "high-quality, low-latency model that strikes a balance between the naturalness of the voice and the generation speed." According to the official documentation, this model supports approximately 32 languages, can handle up to about 40,000 characters in a single request, and offers a low latency performance of approximately 250–300 milliseconds. It is suitable for scenarios that require real-time voice interaction, such as AI voice assistants, dialogue systems, and Agents platforms.
High quality, low-latency model with a good balance of quality and speed.
The Eleven English v2 Speech to Speech (i.e. Voice Changer) model is an English-specific speech-to-speech model developed by ElevenLabs. Its core function is to convert any recorded source audio into another completely identical sound while fully retaining the emotions, intonation, and expression details of the original performance. The official description states that it can capture whispers, laughter, crying, accents, and subtle emotional cues, achieving a highly realistic and nearly human-like effect.
The Eleven Multilingual v2 Speech to Speech (model identifier: eleven_multilingual_sts_v2) is a multilingual speech-to-speech model launched by ElevenLabs. Its core function is: to convert any recorded source audio into another completely identical sound, while fully retaining the emotions, intonation, and expression details of the original performance. The official description states that it can capture whispers, laughter, crying, accents, and subtle emotional cues, achieving a highly realistic and close-to-human effect.
Eleven Multilingual v2 is a high-quality multilingual text-to-speech model provided by ElevenLabs. According to the official documentation, this model is one of the most advanced emotion-aware speech synthesis systems available at present, capable of generating natural, stable and emotionally expressive speech in multiple languages. This model supports approximately 29 languages, including English, Chinese, Japanese, Korean, German, French, Spanish, Hindi, and other major languages. It can maintain consistent speech style and character traits across different languages.
Eleven Multilingual v1 is a multilingual text-to-speech synthesis model (speech generation technology) officially launched by ElevenLabs on April 27, 2023. It is based on the research results of Eleven Monolingual v1 and is developed using richer data, more powerful computing capabilities, and innovative deep learning methods. The core capability of this model is: it can convert text into emotionally rich and highly natural speech in up to 7 new languages.
ElevenLabs Dubbing is an AI dubbing solution launched by ElevenLabs. Its core mission is to create perfect dubs in any language you desire using the original actor's voice, while preserving the original performance.
Scribe v2 is a new generation of speech-to-text model launched by ElevenLabs. It was officially released on January 12, 2026. The official describes it as "the most advanced transcription model", designed specifically for large-scale transcription, captioning, and subtitle production. It further enhances the stability and accuracy on the basis of Scribe v1, and can better handle long audio, pauses, tone changes, and long silences.