ElevenLabs AI Updates: October 31, 2026

1. ElevenLabs Releases SFX v3 Sound Effects Model in Limited Access

ElevenLabs released SFX v3, a text-to-sound-effects model with the ID eleven_text_to_sound_v3, served through the existing sound-generation endpoint by setting the model_id request field; eleven_text_to_sound_v2 remains the default, and the API reference now lists both IDs. ElevenLabs says v3 follows prompts that chain several instructions more reliably (its example is a football crowd cheering, a ball dropping, then the crowd booing), accepts prompts that include music or vocals, produces more natural audio, and clips less on high-frequency sounds such as gongs, chimes, and whistles. The model keeps the existing loop, duration_seconds (0.5 to 30 seconds), and prompt_influence (0 to 1, default 0.3) parameters, and the docs list 40 credits per second when a duration is specified, with no v3-specific pricing. Access is limited and may vary by region; accounts without access receive an “Unknown model” error. SFX v3 is available in ElevenCreative (the Sound Effects tab, Flows, Studio, and the Music Editor), the ElevenLabs mobile app, and the API, and it is documented on the models page but not yet in the changelog. Source