What is Scribe v2?
Scribe v2 is ElevenLabs' current model for transcribing finished recordings, and the replacement for Scribe v1, which ElevenLabs now lists as deprecated. You send an audio or video file of up to 3 GB and 10 hours and get back text with word timestamps, labels for up to 32 speakers and tags for sounds such as laughter. It works out the language itself, even when speakers switch languages mid-file.
It does not handle live audio; ElevenLabs sells Scribe v2 Realtime for that, at $0.39 an hour. Scribe v2 Medical, a fine-tune for clinical audio, costs the same $0.22 and uses the same API.
What changed from Scribe v1?
Scribe v2 replaces Scribe v1, which ElevenLabs now lists as deprecated, and adds controls for jargon and sensitive data.
- Accuracy
- ElevenLabs says it holds up better on long recordings, pauses, changes in tone and long silences.
- Keyterm prompting
- You can pass up to 1,000 words or phrases, such as product names, and the model uses context to decide when they were said.
- Entity detection
- It flags names, card numbers, health details and other sensitive data with timestamps, ready for review or redaction.
- Languages
- It detects every language in a file on its own, so recordings that switch languages need no manual splitting.
Who is Scribe v2 best for?
Scribe v2 is best for developers and teams who transcribe recorded audio in bulk, such as meetings, interviews, calls and videos, especially when the audio mixes languages.
Meetings, interviews and calls
Labels for up to 32 speakers, optional agent and customer roles, and word timestamps turn a multi-person recording into a readable transcript.
Mixed-language audio
It detects each language inside one file, and ElevenLabs rates 36 languages, English, French, German, Spanish and Japanese among them, at 5% word error rate or less.
Recordings full of jargon
Keyterm prompting takes up to 1,000 terms, and entity detection tags 65 kinds of entities such as names, card numbers and health details.
Subtitles and long files in bulk
Files up to 3 GB and 10 hours are accepted, long ones are split and transcribed in parallel, and results can go to a webhook.
What's included with Scribe v2?
Speaker labels and roles, word timestamps and sound tags, keyterm prompting, entity detection and redaction, clean-text options, and webhooks for long jobs.
- Speaker labels and roles
- Labels up to 32 speakers, can tag them as agent and customer for a 10% surcharge, and can match voices saved in a speaker library.
- Timestamps and sound tags
- Word or character timestamps for subtitles, plus tags for non-speech sounds such as laughter or footsteps.
- Keyterm prompting
- Up to 1,000 terms of under 50 characters for $0.05 an hour extra; the model uses context to decide when a term was really said.
- Entity detection and redaction
- Finds 65 kinds of entities, such as names, card numbers and health details, with timestamps, and can swap them for placeholders.
- Clean-text options
- No-verbatim mode drops filler words and false starts, and an experimental editing option applies a plain-language instruction to the finished transcript.
- Webhooks and multichannel audio
- Long jobs can post results to a webhook, and multichannel mode transcribes up to 5 channels with one speaker per channel.
How much does Scribe v2 cost?
Scribe v2 costs $0.22 per hour of audio on the ElevenLabs API; the Free plan includes about 4.5 hours a month, and paid plans from $6 a month include 27 hours or more.
- About 4.5 hours of API transcription a month
- 8 files transcribed at once
- No commercial licence
- 27 hours of API transcription a month
- Extra hours at $0.22 each
- Commercial licence
- 100 hours of API transcription a month
- Extra hours at $0.22 each
- 20 files transcribed at once
- 450 hours of API transcription a month
- Extra hours at $0.22 each
- 40 files transcribed at once
Scale costs $299 a month for 1,359 hours and Business $990 a month for 4,500 hours; Enterprise is priced on request. Pay as you go, without a plan, costs the same $0.22 per hour of audio. Entity detection adds $0.07 an hour and keyterm prompting $0.05 an hour. Paying yearly costs ten months' worth: Starter works out to $5 a month, Creator $18.33 and Pro $82.50. In the ElevenLabs app, transcription draws on the plan's shared credits instead, at 330 credits a minute: 12 minutes a month on Free, about 1.5 hours on Starter and about 6 hours on Creator. Scribe v2 Medical costs the same as Scribe v2. Prices exclude taxes. Full pricing on elevenlabs.io. We checked these prices on .
How do readers rate Scribe v2?
No one has rated Scribe v2 yet. Be the first.
Sign in to rate. One rating per account.
Scribe v2 pros and cons
Users rate its accuracy on non-English and mixed-language audio highly, but it costs more per hour than several newer rivals and cannot transcribe live audio.
Pros
- Developers who compared many models on Hacker News and Reddit single it out for non-English and mixed-language audio, from German, Italian and English meetings to English mixed with Latvian.
- Leads ElevenLabs' own FLEURS test across 30 languages at 95.7% accuracy, ahead of GPT-4o Transcribe at 94.1% and Gemini 2.5 Pro at 93.9%.
- Takes long, busy recordings: up to 10 hours per file with labels for up to 32 speakers, and one tester says it kept speakers apart across an hour of mixed-language audio.
- The hourly price is the same on every plan and on pay as you go, and the Free plan covers about 4.5 hours of API use a month.
Cons
- Costs more per hour than several newer rivals: AssemblyAI's Universal-2 is $0.15 and Mistral's Voxtral Mini Transcribe 2 about $0.18, and several Reddit users call Scribe's pricing high for long meeting recordings.
- Recorded files only: live audio needs Scribe v2 Realtime, which costs $0.39 an hour and does not label speakers.
- Extras add to the bill: keyterm prompting costs $0.05 an hour more and entity detection $0.07, and the Free plan has no commercial licence.
- Accuracy falls off outside the major languages: ElevenLabs rates 19 of them, including Urdu, Khmer and Zulu, at more than 20% word error rate.
What are the best Scribe v2 alternatives?
Three tools on Recaply do Scribe v2's job. Cohere Transcribe is open source, so you can run it on your own server.
AssemblyAIFreemiumChoose it when price per hour matters most: its Universal-2 model transcribes 99 languages for $0.15 an hour, plus $0.02 for speaker labels, with free hours for new accounts.
Voxtral Transcribe 2FreemiumChoose it when your audio is in one of its 13 languages, such as English, Spanish, Hindi or Arabic: batch transcription with speaker labels costs $0.003 a minute, about $0.18 an hour.
Cohere TranscribeOpen SourceChoose it when recordings must stay on your own servers: its open weights run on your own GPUs, though it covers 14 languages against 90+ for Scribe v2.
Questions about Scribe v2
What is the difference between Scribe v2 and Scribe v2 Realtime?
Scribe v2 transcribes finished recordings for $0.22 an hour; Scribe v2 Realtime streams text back within about 150 ms while people speak, for $0.39 an hour. Only Scribe v2 labels speakers, and it accepts up to 1,000 keyterms against 50 for Realtime.
Who is Scribe v2 for?
Scribe v2 is best for developers and teams who transcribe recorded audio in bulk, such as meetings, interviews, calls and videos, especially when the audio mixes languages.
Is Scribe v2 free?
Yes, within limits. The ElevenLabs Free plan includes about 4.5 hours of Scribe v2 transcription through the API each month, or 12 minutes in the ElevenLabs app, but without a commercial licence. Beyond that it costs $0.22 per hour of audio, or a paid plan from $6 a month.
How much does Scribe v2 cost?
Scribe v2's paid plans are Starter at $6 per month, Creator at $22 per month and Pro at $99 per month.
Is there an open source alternative to Scribe v2?
Yes. Cohere Transcribe is an open source tool that does the same job, so you can run your own.
What are the best Scribe v2 alternatives?
The closest alternatives on Recaply are AssemblyAI, Voxtral Transcribe 2 and Cohere Transcribe.
Where can I try Scribe v2?
On its official website, elevenlabs.io.
How long can a Scribe v2 recording be?
Scribe v2 accepts files up to 3 GB and up to 10 hours long. Files longer than 8 minutes are split into as many as four parts that are transcribed in parallel, so a long recording comes back faster than its running time suggests.
Is there a medical version of Scribe v2?
Yes. Scribe v2 Medical is a fine-tune for clinical audio at the same $0.22 an hour, with the same API. ElevenLabs reports about 35% fewer errors than Scribe v2 on clinical dictation, and it is HIPAA-eligible for Enterprise customers with a signed agreement.
Does ElevenLabs keep my Scribe v2 audio and transcripts?
By default, requests are logged so past transcripts stay available. Enterprise customers can turn on zero retention mode, which deletes the audio and the transcript as soon as the request finishes, and can keep data in the EU or India.
Our verdict
Scribe v2 is the model to try first when transcript accuracy matters more than the lowest hourly price, above all for recordings in more than one language.
Developers who tested many models keep picking it for mixed-language meetings and speaker tracking, and it leads ElevenLabs' own FLEURS test at $0.22 an hour.
- Skip it for
- Live audio, which needs Scribe v2 Realtime, and large English-only jobs where a $0.15-an-hour model is good enough.
- Check first
- Run your own recordings through the free 4.5 hours first, and price in the extras if you need keyterms or entity detection.
Scribe v2 Medical
Scribe v2 Realtime
Eleven v4
Eleven v4 Turbo
ElevenLabs Voice Design V3
Gemini 3.8 Flash TTS