Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

ElevenLabs Scribe v3 and Scribe v3 Live

ElevenLabs' November 2026 transcription models: second on Artificial Analysis at 1.9% error, at the same $0.20 an hour as v2.

Reviewed by ReviewRamble editors · Updated

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

4.6/5

Short of a preview model, no transcription you can buy this month makes fewer mistakes, and ElevenLabs did not raise the price to sell it.

What is ElevenLabs Scribe v3 and Scribe v3 Live?

Scribe v3 is ElevenLabs' speech-to-text model for recorded audio, released on 10 November 2026 alongside Scribe v3 Live for real-time streams. Both replaced Scribe v2 as the defaults in the ElevenLabs app and API the same day; switching is a change of model ID to scribe_v3 or scribe_v3_live.

The batch model labels up to 32 speakers, accepts up to 1,000 key terms, tags audio events, detects entities and gives word or character timestamps. New options control whether numbers come out as digits or words and help with audio that mixes languages. The Live model adds turn prediction, a multilingual mode for mid-sentence language switches, real-time speaker labels and a library of up to 10,000 key terms. Both cover more than 90 languages.

The weights are not published. Audio is processed in the US unless you are on an Enterprise plan with an isolated EU, India or Singapore environment.

How much does ElevenLabs Scribe v3 and Scribe v3 Live cost?

PlanPriceWhat you get
Scribe v3$0.20 an hourRecorded audio; key terms +$0.05, editing +$0.07
Scribe v3 Live$0.22 an hourReal-time; key terms +$0.08, editing +$0.12

Scribe v3 $0.20 an hour of audio (key terms +$0.05, transcript editing +$0.07); Scribe v3 Live $0.22 an hour (key terms +$0.08, editing +$0.12). Free tier includes some hours.

Is ElevenLabs Scribe v3 and Scribe v3 Live worth it?

Scribe v3 is a quiet upgrade with a loud result. On Artificial Analysis's independent test of recorded audio, read on 12 November, it scored 1.9% word error rate, second of 64 models and behind only Alibaba's Fun-Realtime-ASR-preview at 1.7%. Its predecessor, Scribe v2, sits at 2.2%. The price stayed at $0.20 an hour, which is the more interesting decision.

ElevenLabs' own headline, 27% fewer errors across 102 languages on the FLEURS set, deserves a footnote the company itself supplies: it averages error per clip rather than across the whole corpus, so its figures (12.38%, down from 16.94%) run higher than other vendors' and cannot be laid beside them. The Live model's numbers are easier to read: on Earnings-22, a set of accented financial calls, error fell from 5.99% to 3.91%.

The catch: speed and geography. Artificial Analysis clocks Scribe v3 at 63 times real time, against 340.8 for Microsoft's MAI-Transcribe-2, which is one tenth of a point behind on accuracy and half the price per minute. Self-serve audio is processed in the US; EU processing needs an Enterprise contract, and Zero Retention Mode is not on by default. The official SDKs also lag the Live model, so turn prediction and the large key-term library need a raw WebSocket.

Our take: for a team already sending audio to ElevenLabs, this is a free improvement with no reason to wait. For a team that cares more about throughput or EU residency than the last tenth of a point, Microsoft and Mistral are now close enough to make the comparison a matter of logistics rather than quality.

Where it earns its place

  • A podcast network publishing in eight languages

    Turn each episode into a clean, speaker-labelled transcript

    Send files to scribe_v3 with diarization and SRT export. A 1,000-hour back catalogue costs about $200 before add-ons.

  • A utility company's customer-service team

    Capture postcodes, account numbers and names correctly on live calls

    Stream calls to scribe_v3_live with a key-term library of street names and product terms; the library is only used when the model hears a likely match. $0.22 an hour plus $0.08 for key terms.

Before you commit

Setup
POST audio to /v1/speech-to-text with model_id scribe_v3, or open the realtime WebSocket with scribe_v3_live. Existing v2 parameters keep working.
Integrations
ElevenLabs API and SDKs, ElevenAgents, webhooks for async results, SRT, DOCX and PDF export. Turn prediction, multilingual mode and the key-term library currently need a direct WebSocket.
Data and privacy
No published weights; no self-hosting. Self-serve audio is processed and stored in the US. EU, India and Singapore isolated environments are Enterprise-only, and Zero Retention Mode is not on by default.
Cost at scale
10,000 hours a year cost about $2,000 on Scribe v3 before add-ons; silent stretches are billed. Artificial Analysis lists it at $3.33 per 1,000 minutes.
Limits
Concurrency runs from 8 batch and 6 real-time jobs on Free to 60 and 45 on Scale or Business. Maximum file length is not stated on the models page.

What’s good

  • 1.9% error on Artificial Analysis, second of 64 models
  • Same price as Scribe v2: $0.20 an hour batch, $0.22 live
  • Up to 32 speakers, 1,000 key terms and audio-event tags
  • Upgrade is a one-line model ID change

What’s not

  • About five times slower than MAI-Transcribe-2 on Artificial Analysis
  • US processing unless you buy Enterprise residency
  • Own FLEURS figures are not comparable with other vendors'
  • SDKs do not yet support Live turn prediction or the key-term library

Best for

Product teams transcribing calls, meetings or media in many languages who want top-tier accuracy without a price change or a new integration.

ElevenLabs Scribe v3 and Scribe v3 Live alternatives

ElevenLabs Scribe v3 and Scribe v3 Live: common questions

Is ElevenLabs Scribe v3 and Scribe v3 Live worth it?
Short of a preview model, no transcription you can buy this month makes fewer mistakes, and ElevenLabs did not raise the price to sell it. We rate it 4.6 out of 5.
How much does ElevenLabs Scribe v3 and Scribe v3 Live cost?
Scribe v3 $0.20 an hour of audio (key terms +$0.05, transcript editing +$0.07); Scribe v3 Live $0.22 an hour (key terms +$0.08, editing +$0.12). Free tier includes some hours.
What are the downsides of ElevenLabs Scribe v3 and Scribe v3 Live?
About five times slower than MAI-Transcribe-2 on Artificial Analysis; uS processing unless you buy Enterprise residency; own FLEURS figures are not comparable with other vendors'; sDKs do not yet support Live turn prediction or the key-term library.
Who is ElevenLabs Scribe v3 and Scribe v3 Live best for?
Product teams transcribing calls, meetings or media in many languages who want top-tier accuracy without a price change or a new integration.
How much does ElevenLabs Scribe v3 and Scribe v3 Live cost for a team or in production?
10,000 hours a year cost about $2,000 on Scribe v3 before add-ons; silent stretches are billed. Artificial Analysis lists it at $3.33 per 1,000 minutes.
Is ElevenLabs Scribe v3 and Scribe v3 Live safe for company data?
No published weights; no self-hosting. Self-serve audio is processed and stored in the US. EU, India and Singapore isolated environments are Enterprise-only, and Zero Retention Mode is not on by default.
What are the best ElevenLabs Scribe v3 and Scribe v3 Live alternatives?
The closest alternatives we have reviewed are ElevenLabs Scribe v2, MAI-Transcribe-2-Streaming, Voxtral 3 (Transcribe and TTS) and AssemblyAI STT-3.7 (batch, Realtime and Realtime-Agent).

Answers checked Nov 21, 2026.

Make ElevenLabs Scribe v3 and Scribe v3 Live? Show this rating on your siteElevenLabs Scribe v3 and Scribe v3 Live rated 4.6 out of 5 on ReviewRamble