Transcription · ElevenLabs
Scribe
ElevenLabs' transcription model, covering over 90 languages. Anonymized, and the dearest transcription here.
Anonymized Moderate costWhat sets it apart
Scribe is ElevenLabs' transcription model with the lowest word error rate on the Artificial Analysis test among these four. It labels speakers and non-speech sounds, detects entities and takes up to 100 key terms to spell correctly, at a higher price.
- Over 90 languages, mixed in one file
- Keyterm prompting for up to 100 terms
- Entity detection and sound tags
Recommended for: transcribe several languages or strong accents.
Strengths and limits
Where it shines
- Over 90 languages with automatic detection
- Labels non-speech sounds, per ElevenLabs
Think twice when
- Anonymized, not private: ElevenLabs' retention policy applies
- Costs about 1.6 times Parakeet or Whisper
- 90+ languages
Use cases
Transcribe a panel with several speakers
Transcribe this panel discussion, label each speaker, and note laughter or applause in brackets.
It labels speakers and tags non-speech sounds.
Get product names spelled correctly
Transcribe this sales call. Spell correctly: Sub Rosa, Carpe Diem, Kubernetes, Aurélien Dufresne.
Keyterm prompting steers rare names and jargon.
Against its rivals
- WhisperWhisper is private and cheaper, with a higher error rate on the Artificial Analysis test.#3 of 4 on Artificial Analysis word error rate (4.1%)
- ParakeetParakeet is private and cheaper but covers 25 languages.
- xAI Speech to TextxAI is cheaper, with similar accuracy on that test.#2 of 4 on Artificial Analysis word error rate (4%)
Version history
Each version, dated, with what it changed over the one before.
- Scribe v1Earlier, not in Sub Rosa
- 99 languages, speaker diarization, word timestamps and audio event tagging
- Scribe v2Latest
- Better stability on long audio, pauses and silences
- Keyterm prompting (up to 100 terms), entity detection and automatic mixed-language transcription
Benchmarks
How to read these scores: what a benchmark tells you.
Artificial Analysis word error rate: #1 of 4 models measured in the catalog. The share of words a transcription gets wrong (missed, added or replaced) on a mixed set of real recordings. Lower is better. At 5%, one word in twenty needs fixing; under 3% a transcript reads almost clean.
See the numbers
| Model | % |
|---|---|
| 1. ElevenLabs Scribe V2 | 2.2% |
| 2. xAI Speech to Text v1 | 4% |
| 3. Whisper Large V3 | 4.1% |
| 4. Wizper (Whisper v3) | 4.7% |
The model in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
ElevenLabs Scribe V2elevenlabs/scribe-v2 | Transcription | Anonymized | $0.27 per hour of audio |
Specifications
| Specification | Scribe v2 |
|---|---|
| Released | January 9, 2026 |
| Weights | Closed |
| Accepts | audio |
| Produces | text |
| Languages | 90+ languages |
| Price | $0.27 per hour of audio |
| Privacy | Anonymized |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.