Short answer: you can’t run an ElevenLabs Agent on the v3 TTS model today.
v3 (Alpha) is not suitable for real-time/conversational use and is intended for offline-style generations; use Turbo v2.5 or Flash v2.5 for agents. (help.elevenlabs.io)
WebSocket streaming (what Agents use for low-latency speech) is not available for the eleven_v3 model. (elevenlabs.io)
In Agent settings, the selectable “Model family” options are Flash, Turbo, or Multilingual—not v3. (elevenlabs.io)
What you can do instead:
If you want the timbre/character of a v3-designed voice, create it in Voice Design v3 and then use that saved voice with a supported model family (Flash/Turbo/Multilingual) in your agent; v3-designed voices are backward-compatible with v2-family models. (elevenlabs.io)
For non-real-time workflows outside Agents, you can generate with v3 by calling the Create/Stream speech endpoints with model_id=eleven_v3. (help.elevenlabs.io)
If/when real-time v3 becomes supported in Agents, ElevenLabs will document it in the model/agents docs; for now, stick to Flash/Turbo for live agents. (elevenlabs.io)
Via REST API (Text to Speech): set model_id to eleven_turbo_v2_5 when calling Create speech or Stream speech. The model_id parameter is optional and defaults to another model, so you must override it. See API reference. (elevenlabs.io)
Via WebSocket streaming (real‑time TTS): include model_id=eleven_turbo_v2_5 in the stream-input URL query, e.g. wss://api.elevenlabs.io/v1/text-to-speech/{voice_id}/stream-input?model_id=eleven_turbo_v2_5. The docs show this pattern; swap in eleven_turbo_v2_5. (elevenlabs.io)
In Agents (UI): in your agent settings, under Voice > Multi‑voice support, add/edit a voice and set Model Family to Turbo. This routes the agent’s TTS through the Turbo family. (elevenlabs.io)
Useful references:
Turbo v2.5 model ID and overview: eleven_turbo_v2_5. (elevenlabs.io)
Listing available models (GET /v1/models) and model IDs. (help.elevenlabs.io)
Character limit for Turbo v2.5 (40,000 chars per request). (elevenlabs.io)
If you need to enforce a language during streaming, the Stream speech API supports a language_code parameter and notes that Turbo v2.5 supports language enforcement. (elevenlabs.io)
regal barn
restive lake