ElevenLabs Rolls Out Scribe v3 as Its Default Transcription Model

On November 10, 2026, ElevenLabs launched two new speech-to-text models: Scribe v3, built for recorded audio such as meeting recordings, interviews, and call archives, and Scribe v3 Live, a low-latency model designed for live calls and voice agent use. Coverage from developers and the press followed over the following days, with the first reports appearing on November 12.

Writing for The Decoder on November 12 at 21:48, the outlet reported that the two models are positioned around accuracy in difficult conditions, specifically transcribe reliably through background noise and crosstalk. It noted that Scribe v3 Live also draws on conversational context to stop a voice agent from interrupting the caller. Citing benchmarks from Artificial Analysis, the article stated that Scribe v3 narrowly edges out competitors from OpenAI, Google, and Microsoft on accuracy, but added that rivals such as Microsoft and xAI run many times faster while charging roughly half the price.

In a piece published on DEV Community on November 13 at 02:00, it was reported that Scribe v3 has been set as the default transcription model inside ElevenLabs applications. According to the company, Scribe v3 lowers word errors by 27% compared with v2 across 102 languages, and reduces speaker-diarization errors by 17% on recordings with three or more speakers. Developers can switch to the new model by changing a single model ID line. Both models are available through ElevenAPI and related channels, and the story currently stands at the stage of release and early developer adoption, with the accuracy-versus-speed-and-cost tradeoff still open between ElevenLabs and its rivals.