What it is
Scribe v2 is ElevenLabs' batch speech-to-text model, launched on 9 January 2026, two months after the low-latency Scribe v2 Realtime. It covers 90+ languages, handles several languages in one file, and adds keyterm prompting, diarization for up to 32 speakers, detection of 65 entity types with timestamps, and audio event tagging.
Scribe v2 Medical, tuned for clinical audio, is HIPAA-eligible under a BAA and, by ElevenLabs' account, makes 35% fewer errors on clinical audio than the base model. Scribe v2 Realtime targets voice agents at about 150 ms latency.
Our review
Scribe v2 is among the most accurate transcription models available, but it is no longer the obvious pick on accuracy and price together. On Artificial Analysis's word error rate index (AA-WER) it scores 2.2%, fourth on the board, behind Alibaba's Fun-Realtime-ASR preview (1.7%), Microsoft's MAI-Transcribe-2 (2.0%) and Mistral's Voxtral Small Transcribe 3 (2.1%). It still beats Google's Gemini 3.5 Transcribe (2.6%) and OpenAI's GPT Transcribe (3.3%).
MAI-Transcribe-2 is the uncomfortable comparison. Artificial Analysis lists it at $1.67 per 1,000 minutes and 307 times real-time, against Scribe v2's $3.67 and 69 times: better accuracy at less than half the price and over four times the speed. Deepgram Nova-3 is faster again at 546 times real-time, but its 5.2% error rate is more than double Scribe's.
What keeps Scribe in contention is the work around the words. Diarization for up to 32 speakers, entity detection with timestamps for redaction, up to 1,000 keyterms per batch request and a HIPAA-eligible medical variant are features a meeting or clinical product would otherwise build itself. Keyterms cost an extra $0.05 an hour, and silence is billed like speech.
Check the data defaults before sending recordings. ElevenLabs keeps request history by default, deleted items can sit in backups for up to 30 days, and Zero Retention Mode, switched on per request with enable_logging=false, is offered only to selected enterprise customers. The trust centre lists SOC 2 Type II, ISO 27001 and HIPAA, so regulated buyers can get there, but only through a sales contract.
Where it earns its place
A meeting-notes start-up transcribing 20,000 hours of calls a month
Get speaker-labelled transcripts with names and jargon right
Send recordings with keyterms for product and customer names: 20,000 hours costs about $4,400, or $5,400 with keyterms on. Files over 8 minutes are split into up to four parallel chunks, which shortens turnaround.
A hospital digital team piloting ambient clinical notes
Transcribe consultations without storing patient audio
Use Scribe v2 Medical under a BAA with Zero Retention Mode; ElevenLabs claims 35% fewer clinical errors than base Scribe v2. Both require an enterprise agreement through sales, so volume terms are negotiated.
A compliance engineer at a contact centre
Find and redact card numbers and personal data in recorded calls
Turn on entity detection, which returns timestamps for 65 entity types including card numbers, then mute those spans in the audio. At $0.22 an hour, 1,000 hours of calls a day is about $6,600 a month.
Before you commit
- Setup
- An API key and one POST to /v1/speech-to-text with the file; long jobs can deliver results to a webhook set in workspace settings.
- Integrations
- ElevenLabs REST API and SDKs, webhooks for async results, ElevenCreative and ElevenAgents; Scribe v2 Realtime for live use.
- Data and privacy
- Request history is kept by default; Zero Retention Mode is for selected enterprise customers. The trust centre lists SOC 2 Type II, ISO 27001, ISO 42001, GDPR, HIPAA and PCI, and ElevenLabs offers EU and India data residency. Training use of API audio is not addressed on the pages we read.
- Cost at scale
- $0.22 per audio hour on every plan, so 10,000 hours a month is $2,200, plus $500 with keyterms. The $990 Business plan includes 4,500 hours; Realtime is $0.39 an hour.
- Limits
- 3 GB and 10 hours per file, up to 5 channels in multichannel mode, and 1,000 keyterms per batch request (50 in Realtime). Batch speed is about 69 times real-time on Artificial Analysis's test.
What's good
- 2.2% AA-WER, fourth on Artificial Analysis's board
- Diarization for up to 32 speakers
- 65 entity types with timestamps for redaction
- HIPAA-eligible medical variant at the same $0.22 per hour
- Files up to 3 GB and 10 hours
What's not
- MAI-Transcribe-2 is more accurate, cheaper and faster on Artificial Analysis's tests
- 69x real-time, far slower than Deepgram Nova-3
- Zero Retention Mode is enterprise-only
- Silence is billed
Best for
Meeting, media and clinical products that need speaker labels and redaction as well as an accurate transcript.
Alternatives
ElevenLabs Scribe v2: common questions
- Is ElevenLabs Scribe v2 worth it?
- Scribe v2 transcribes accurately and does the redaction and speaker work for you, but Microsoft now sells better accuracy for less than half the price. We rate it 4.4 out of 5.
- How much does ElevenLabs Scribe v2 cost?
- API: Scribe v2 and v2 Medical $0.22 per hour of audio; Scribe v2 Realtime $0.39 per hour; keyterm prompting adds $0.05 per hour. Plans from $6/month include hours.
- What are the downsides of ElevenLabs Scribe v2?
- MAI-Transcribe-2 is more accurate, cheaper and faster on Artificial Analysis's tests; 69x real-time, far slower than Deepgram Nova-3; zero Retention Mode is enterprise-only; silence is billed.
- Who is ElevenLabs Scribe v2 best for?
- Meeting, media and clinical products that need speaker labels and redaction as well as an accurate transcript.
- How much does ElevenLabs Scribe v2 cost for a team or in production?
- $0.22 per audio hour on every plan, so 10,000 hours a month is $2,200, plus $500 with keyterms. The $990 Business plan includes 4,500 hours; Realtime is $0.39 an hour.
- Is ElevenLabs Scribe v2 safe for company data?
- Request history is kept by default; Zero Retention Mode is for selected enterprise customers. The trust centre lists SOC 2 Type II, ISO 27001, ISO 42001, GDPR, HIPAA and PCI, and ElevenLabs offers EU and India data residency. Training use of API audio is not addressed on the pages we read.
- What are the best ElevenLabs Scribe v2 alternatives?
- The closest alternatives we have reviewed are Gemini 3.5 Transcribe, GPT Transcribe, Deepgram Nova-3 and Whisper large-v3-turbo.
Answers checked Oct 23, 2026.
Make ElevenLabs Scribe v2? Show this rating on your site
