Scribe v2 Realtime

What is Scribe v2 Realtime?

Scribe v2 Realtime is ElevenLabs' current model for live transcription, the streaming partner of the batch Scribe v2. You stream audio over a WebSocket and get partial text in about 150 ms, then final segments that you commit yourself or leave to voice activity detection at pauses. It guesses likely next words and punctuation to keep the delay down, and picks up where it left off after a dropped connection.

It is offered through the API and inside ElevenLabs Agents, where it is an option rather than the default listening model. It does not label speakers, which ElevenLabs says is not a priority for this model.

Should I use Scribe v2 Realtime or Scribe v2?

Use Realtime only when text must appear while people are still talking; for recordings, Scribe v2 costs less and does more.

Timing

Realtime returns partial text in about 150 ms; Scribe v2 works on finished files of up to 10 hours.

Price

$0.39 an hour for Realtime against $0.22 for Scribe v2.

Speakers

Scribe v2 labels up to 32 speakers; Realtime labels none.

Keyterms

Realtime takes 50 keyterms of up to 20 characters; Scribe v2 takes 1,000 of up to 50.

Who is Scribe v2 Realtime best for?

Scribe v2 Realtime is best for developers building voice agents, live captions and meeting tools that need text within a fraction of a second of speech.

• Voice agents

  Partial transcripts arrive in about 150 ms, and ElevenLabs Agents can use it as its listening model.

• Live captions and meeting notes

  Audio streams in and text streams back, with voice activity detection splitting speech at pauses.

• Phone and call audio

  It takes PCM audio from 8 to 48 kHz and μ-law encoding, which covers telephony as well as browser and studio sources.

• Live apps in many languages

  It transcribes 90+ languages and recognises the language on its own, even when speakers switch mid-conversation.

What's included with Scribe v2 Realtime?

Partial and final text, voice activity detection or manual commits, keyterm prompting, entity detection, optional word timestamps, a background-noise filter and single-use tokens for browser apps.

Partial and final text

Draft text streams back while people speak, followed by committed segments that will not change.

Voice activity detection or manual commits

Segments close on their own at pauses, with adjustable thresholds, or whenever your app decides, for example on a turn-taking signal.

Keyterm prompting

Up to 50 terms of 20 characters each, such as product or drug names, for a 20% surcharge on the hourly rate.

Entity detection

Flags personal, health and payment details in final segments, with their positions in the text.

Timestamps and language detection

Optional follow-up messages give word-level timing and the detected language after each commit.

Background audio filter

Reduces false triggers from nearby conversations and room noise, though it cannot be combined with timestamps.

Single-use tokens

Tokens made on your server and valid for 15 minutes let a web or mobile app stream audio without exposing your API key.

How much does Scribe v2 Realtime cost?

Scribe v2 Realtime costs $0.39 per hour of audio on the ElevenLabs API; the Free plan includes about 2.5 hours a month, and paid plans from $6 a month include 15 hours or more.

Scale costs $299 a month for 767 hours and Business $990 a month for 2,538 hours, both with 45 live streams at once; Enterprise is priced on request with higher limits. Pay as you go, without a plan, costs the same $0.39 per hour of audio. Keyterm prompting adds 20% to the hourly rate. Paying yearly costs ten months' worth: Starter works out to $5 a month, Creator $18.33 and Pro $82.50, with the same included hours. The API bills these in US dollars rather than app credits, and prices exclude taxes. Full pricing on elevenlabs.io. We checked these prices on 28 Sep 2026.

How do readers rate Scribe v2 Realtime?

No one has rated Scribe v2 Realtime yet. Be the first.

Scribe v2 Realtime pros and cons

It is fast, multilingual and ready for phone audio, but it costs more than batch Scribe v2 and cannot tell speakers apart.

What are the best Scribe v2 Realtime alternatives?

Scribe v2 for recorded audio, AssemblyAI for cheaper English streaming or live speaker labels, and Voxtral Transcribe 2 for open weights you can run yourself.

Scribe v2Free

Choose it when the audio is already recorded: it costs $0.22 an hour instead of $0.39 and labels up to 32 speakers.

AssemblyAIFreemium

Choose it when you need speaker labels on live audio, a $0.12-an-hour add-on, or English-only streaming for $0.15 an hour with Universal-Streaming.

Voxtral Transcribe 2Freemium

Choose it when you want to run live transcription on your own hardware: Voxtral Realtime's weights are open under Apache 2.0, and it covers 13 languages.

Questions about Scribe v2 Realtime