Short answer: It depends what you mean by “v3 voices.”
If you mean the Eleven v3 TTS model: No. Eleven v3 (Alpha) isn’t made for real‑time use and isn’t supported on Agents; use Flash v2.5, Turbo v2.5, or Multilingual v2 for Agents. Eleven’s docs explicitly say v3 is not for real‑time/Agents, and list the supported models for Agents. (elevenlabs.io)
If you mean voices created with Voice Design v3 (i.e., the “v3 voice” assets): Yes. Those voice profiles are backward‑compatible with the v2 family, so you can select them for Agents; they’ll render using the Agent‑supported models (Flash/Turbo/Multilingual), not the v3 model itself. (elevenlabs.io)
Additional reference on v3‑format voices being backward compatible: Voice Remixing outputs a v3 voice that remains compatible with other models. (elevenlabs.io)
hasty ravine
ember heron