New eleven_v4 model: more expressive speech and better voice cloning in 90+ languages. eleven_v4_turbo answers in about 100 ms.
On September 28, 2026 ElevenLabs released two new speech models. eleven_v4 gives higher-quality, more expressive speech and improved voice cloning across more than 90 languages. eleven_v4_turbo is the same family for real-time use, with a median response time of about 100 ms.
Wirestack
ELEVENLABS_API_KEY (pass it to the SDK as api_key; over plain HTTP it goes in the xi-api-key header)elevenlabs for Python or @elevenlabs/elevenlabs-js for TypeScripteleven_v4 through the Text to Dialogue API, for content, audiobooks and long-form audio.eleven_v4_turbo through the Text to Dialogue WebSocket, for agents and anything interactive.eleven_v3, so you have to name the new model yourself:from elevenlabs import ElevenLabs, DialogueInput
client = ElevenLabs(
api_key="xi-api-key",
)
client.text_to_dialogue.convert(
model_id="eleven_v4",
inputs=[
DialogueInput(
text="[giggling] Knock knock",
voice_id="JBFqnCBsd6RMkjVDRZzb",
),
DialogueInput(
text="[curious] Who is there?",
voice_id="Aw4FAjKCGjjNkVhN1Xmq",
)
],
)Run the script with your own key and two voice IDs from your voice library. The call returns the generated audio; listen for the tags in square brackets ([giggling], [curious]) being acted out and not read aloud.
model_id="eleven_v4" added, as the changelog instructs. The reference page does not show a complete sample with that model set.