Here are the Text-to-Speech parameter differences that are explicitly called out in the official docs:
use_speaker_boost (voice_settings): Not available for eleven_v3. It remains available for eleven_multilingual_v2 and eleven_flash_v2_5. Source: “Speaker Boost is not available for the Eleven v3 model.” (elevenlabs.io)
SSML break tags: Not supported by eleven_v3; use audio tags instead. Other models (e.g., multilingual_v2, flash_v2_5) can use SSML breaks (and SSML parsing can be enabled in WebSocket flows). (elevenlabs.io)
WebSocket TTS parameters (enable_ssml_parsing, auto_mode, sync_alignment, inactivity_timeout, etc.): Not applicable to eleven_v3 because TTS WebSockets are not available for this model; they remain available for multilingual_v2 and flash_v2_5. (elevenlabs.io)
apply_text_normalization: All three models accept the parameter, but flash_v2_5 restricts turning it “on” (Enterprise plans only). For flash_v2_5 you typically must use auto/off; multilingual_v2 and eleven_v3 allow on/off/auto. (elevenlabs.io)
Deprecated (global, not v3-specific): use_pvc_as_ivc is marked Deprecated on Text-to-Speech requests. Applies across models (avoid relying on it for v3, multilingual_v2, or flash_v2_5). (elevenlabs.io)
If you need more detail on model capabilities and selection, see the models overview. (elevenlabs.io)
winged vessel
hasty carbon