ElevenLabs launches Scribe v3 with a separate live transcription model

ElevenLabs introduced Scribe v3 and v3 Live on November 10. The former targets recorded audio; the latter adds real-time speaker labels and conversational turn detection. [1]

The company also describes terminology and formatting controls. Its accuracy improvements are provider-reported, not results independently tested by SpicyRanks. [1]

ElevenLabs' methodology distinguishes word errors, semantic errors and speaker-label errors. It warns that its live and batch diarization scores are not directly comparable. [1]

Spicy's takeTreat that distinction as a testing requirement: check words, speaker attribution and turn-taking separately on representative recordings. [1]

Spicy's takeThe pricing page says products draw from one monthly credit pool and gives approximate speech-to-text usage of 330 credits per minute. It does not identify a universal v3/v3 Live API rate; confirm the applicable product and plan before estimating costs. [2]