Scribe V2 Medical Generally Available

ElevenLabs reports that Scribe v2 Medical, a speech-to-text model adapted for clinical audio, made fewer transcription errors than base Scribe v2 on dictated notes. Its release article reports 4.9% overall word error rate versus 7.6% on Corti's MedDictate test in English, French and German. Word error rate counts inserted, missing and substituted words relative to a reference transcript.

On the separate Eka benchmark, ElevenLabs reports 14.3% error for isolated medical terms versus 7.5% for terms within sentences. That contrast matters when interpreting the improvement on dictated notes: a drug name spoken alone remains a different and harder test. These are company-reported transcription benchmarks, not measures of clinical outcomes or a guarantee of clinical accuracy.