mishal brands

Heads up: some links are affiliate links. If you sign up through one, we may earn a commission. It costs you nothing. Read the affiliate disclosure.

Trend · AI transcription

ElevenLabs Scribe v3: transcripts that name the right speaker

Scribe v3 is ElevenLabs’ new speech-to-text model for recorded audio, released November 10, 2026. ElevenLabs reports 27% fewer word errors than Scribe v2 across 102 languages and 17% fewer speaker-label errors on recordings with three or more speakers. A live version, Scribe v3 Live, brings speaker labels to realtime transcription.

Last checked · Announced · By Mishal Brands Editorial

Checked Nov 10, 2026. Everything below is what the company reports in its own announcement, linked at the bottom: ElevenLabs, “Introducing Scribe v3 and v3 Live, our most accurate speech to text models” (Nov 10, 2026). We have not tested it ourselves, and figures are the company’s own unless we say otherwise.

What it does and why it matters

Every transcript error travels downstream into notes, summaries and agent decisions. ElevenLabs built Scribe v3 for messy, real audio: several people talking over each other, a caller who pauses mid-sentence, or speakers switching languages.

ElevenLabs reports that Scribe v3 makes 27% fewer word errors than Scribe v2 across 102 languages, and 17% fewer diarization (speaker-label) errors on recordings with three or more speakers. Scribe v3 Live, the low-latency version, makes 43% fewer errors than Scribe v2 Realtime on semantic word error rate and adds speaker labels to realtime transcription for the first time (ElevenLabs, Nov 10, 2026).

ElevenLabs puts the gap in plain numbers: a one-hour meeting at 150 words a minute is about 9,000 words, so going from a 4% to a 2% error rate is 180 fewer wrong words.

Who it is for

  • Teams that record meetings, interviews or podcasts and need to know who said what.
  • Builders of voice agents who need callers’ numbers and names heard right the first time.
  • Multilingual teams, including Hindi and English speakers who switch mid-sentence (Scribe v3 Live adds a multilingual mode).
  • Healthcare and regulated teams: ElevenLabs says both models are HIPAA-eligible with a Business Associate Agreement and offers a Zero Retention Mode.

Standout features

  • Better speaker labels. Lowest diarization error of any model ElevenLabs tested, with 12% fewer errors than Scribe v2 overall.
  • Realtime speaker labels. On the English part of ElevenLabs’ real-world dataset, Scribe v3 Live’s diarization error rate is 3.1%, against 5.9% to 7.3% for the other realtime models it tested.
  • Turn detection that listens. Scribe v3 Live uses conversation context, not just silence, so a caller who pauses halfway through a number is not cut off.
  • Keyterm prompting. Give it up to 1,000 terms such as brand names and jargon; Live adds a keyterm library for vocabularies too big for a list.
  • Formatting in plain English. Choose digits or words for numbers and write rules like “write every date as YYYY-MM-DD.”
  • Data residency options. Runs in the US by default, with isolated EU, India and Singapore environments for enterprise customers.

How to start

  1. Open the ElevenLabs app: ElevenLabs says Scribe v3 and v3 Live are now the default speech-to-text models there.
  2. Upload one real recording with three or more speakers and check the speaker labels first.
  3. Add your brand names, product names and acronyms as keyterms before you transcribe a batch.
  4. Developers on the API switch by changing the model ID to scribe_v3 or scribe_v3_live.
  5. Building a voice agent? Pick Scribe v3 Live in ElevenAgents and test with callers reading out numbers.

Alternatives

  • Otter.ai, Fireflies.ai and Fathom if you want a meeting notetaker app rather than a transcription model.
  • Descript if you want to edit audio and video by editing the transcript.
  • Sembly and tl;dv for meeting notes with summaries and CRM sync.

The catches

  • All accuracy figures are ElevenLabs’ own benchmarks. ElevenLabs says it scored each language separately before averaging, so its numbers are not directly comparable with results measured another way.
  • The real-world diarization dataset is ElevenLabs’ own; it says it will open source it “soon.”

Frequently asked questions

What is ElevenLabs Scribe v3?

ElevenLabs’ newest speech-to-text model for recorded audio, released November 10, 2026. ElevenLabs reports 27% fewer word errors than Scribe v2 across 102 languages and the most accurate speaker labels of any model it tested.

What is the difference between Scribe v3 and Scribe v3 Live?

Scribe v3 is for recorded audio. Scribe v3 Live is the low-latency version for realtime audio, used in voice agents; it adds speaker labels and context-aware turn detection to realtime transcription.

Is Scribe v3 free to try?

ElevenLabs’ pricing page lists a free plan with monthly credits, and Scribe v3 is the default speech-to-text model in the ElevenLabs app. Check the pricing page for how many credits transcription uses on your plan.

Pricing and free plan for ElevenLabs

The Scribe v3 announcement does not list new prices. It is available in the ElevenLabs app and through the API on your existing plan.

Plan facts from our library, worded as each vendor reports them:

Related on Mishal Brands

Sources (checked Nov 10, 2026)

Browse by category: All 153 Business and productivity tools. Every tool page lists the vendor’s own price and the date we checked it.