Launching today

Scribe v3 Live Medical
ElevenLabs’ medical transcription model built for agents
1 follower
ElevenLabs’ medical transcription model built for agents
1 follower
Scribe v3 Live Medical is ElevenLabs’ real-time speech-to-text model tuned for clinical conversations and built for healthcare voice agents. Per ElevenLabs, it scored 12.8% WER on PriMock57 and cut medical term errors by up to 48% with up to 10,000 custom terms. It waits for callers to finish reading IDs and dates, follows mid-call language switches, and matches Scribe v3 Live’s roughly 150 ms latency. HIPAA-eligible for enterprise with a BAA. Available now in ElevenAgents.




Launch tags:Artificial Intelligence
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
ElevenLabs just released Scribe v3 Live Medical, a real-time speech-to-text model tuned for clinical conversations. It’s built on Scribe v3 Live, and ElevenLabs calls it its first medical transcription model built for agents.
What it adds for healthcare voice agents:
Custom terms: load up to 10,000 terms, which the model only uses when a caller seems to say one. In ElevenLabs’ tests this cut errors on medical terms by up to 48%.
Turn detection: the agent waits for callers to finish reading numbers like member IDs or dates of birth.
Language switching: it follows mid-call switches between languages instead of dropping the second one.
Per ElevenLabs’ internal tests:
12.8% word error rate on PriMock57 (mock primary care consultations), the lowest among the medical models tested and 17% fewer errors than the next closest
0.6% semantic WER on Fareez OSCE (patient interviews), also the lowest among the medical models tested
Latency matches Scribe v3 Live at about 150 ms
It supports 90+ languages, though medical terminology was tested in English, French and German, and most benchmarks are in English. On everyday speech it performs about the same as Scribe v3 Live, so the gains are specific to medical language.
For data handling, ElevenLabs says it’s HIPAA-eligible for enterprise customers with a business associate agreement and Zero Retention Mode. Patient data is stored in the US by default, with isolated EU, India or Singapore environments for enterprise customers.
It’s available now in ElevenAgents. To enable it, go to Settings > Automatic Speech Recognition and pick it from the model dropdown.
Good fit if you’re building healthcare voice agents for intake, scheduling or patient calls and want better accuracy on drug names and clinical terms. The accuracy numbers are ElevenLabs’ own, so it’s worth testing on your own calls.
Learn more: ElevenLabs announcement