Scribe v3 Live Medical
Intro
Scribe v3 Live Medical is the real-time medical speech-to-text model ElevenLabs released on 17 November 2026, in a post by John Durovsik. Officially, it is built on the existing real-time model Scribe v3 Live, optimized for clinical language, and is available now in ElevenAgents. The medical transcription product page describes two Speech to Text models for clinical language: Scribe v3 Live Medical for ElevenAgents, and Scribe v2 Medical for batch transcription.
The stated motivation is concrete: a voice agent can only act on what it hears, and drug names are easy to mishear. In ElevenLabs’ testing, a speech model transcribed sitagliptin (a diabetes medication) as citalopram (an antidepressant). The company already offered medical transcription for recorded audio such as dictation and visit recordings, but scheduling, medical information lines, eligibility checks, and claims follow-ups happen on live calls that cannot wait for a recording to end. The post calls Scribe v3 Live Medical “our first medical transcription built for agents,” carrying the clinical focus of Scribe v2 Medical into real-time conversations.
This site already filed the general model Scribe v3 / Scribe v3 Live on 11 November 2026. This entry is not a repeat of that launch, and it should not be described as a routine parameter tweak of Scribe v3 Live: ElevenLabs presents it as a distinct model name for clinical language on live calls.
Advertisement
Explain
Officially listed capabilities for clinical calls include loading up to 10,000 custom terms, which the model draws on only when a caller seems to say one of them; with a custom list loaded, ElevenLabs reports up to 48% fewer errors on medical terms in its tests. The model is also described as recognizing when a number is not finished yet, so an agent can wait through pauses in a member ID or date of birth, and as supporting more than 90 languages so mid-call language switches are transcribed instead of dropped.
On evaluation, the post says the results come from ElevenLabs’ internal testing of selected providers in real-time mode and do not include every available model. On PriMock57 (57 unscripted mock live primary-care consultations of about nine minutes each, with phone audio and background noise), Scribe v3 Live Medical had a word error rate of 12.8%, the lowest among the medical models tested and about 17% fewer errors than the next closest one. On Fareez OSCE (272 recorded patient interviews), it had the lowest semantic WER among the medical models tested, at 0.6%. Compared with Scribe v3 Live on Eka Medical ASR, which focuses on medical terms and branded drug names, its semantic WER was 11% versus 11.54% (4.7% fewer errors). On the English everyday sets of Common Voice and FLEURS, it performed about the same as Scribe v3 Live. On MED-Term, semantic WER was 2.24% in English, 2.81% in German, and 3.24% in French. On latency, the post says Scribe v3 Live transcribes speech in about 150 milliseconds, and Scribe v3 Live Medical matched that latency in testing.
On compliance, ElevenLabs says Scribe v3 Live Medical is HIPAA-eligible for enterprise customers with a business associate agreement (BAA) in place and Zero Retention Mode enabled. It runs on ElevenAgents, which is covered by SOC 2 Type II, ISO 27001, and PCI DSS Level 1. Patient data is stored in the US by default; enterprise teams that need storage outside the US can use isolated environments in the EU, India, or Singapore. The post also quotes early users at Pfizer (Idin Jafari) and Vintage Labs AI (Shantanu Ghorai); those are customer testimonials, not third-party benchmarks.
Usage
Official getting-started steps: Scribe v3 Live Medical is available today in ElevenAgents. Go to Settings > Automatic Speech Recognition and select it from the model dropdown. Product and compliance details are on the medical transcription page and in the blog post.
It is aimed at patient, member, and provider calls where clinical details such as medications, doses, procedures, or symptoms come up. ElevenLabs also says it holds its accuracy on everyday speech compared with Scribe v3 Live, so teams can use it on the same line for non-clinical calls such as rescheduling or asking about office hours. If a team already transcribes recordings or dictation, batch work should still use Scribe v2 Medical as described on the product page, rather than treating Live Medical as an offline batch model. For in-house drug, clinic, or provider names, load up to 10,000 custom terms as the docs describe. When calls include protected health information, enterprise customers need to meet the official conditions, including a BAA and Zero Retention Mode.