Tasks and inputs
Scribe v3 is listed for Automatic speech recognition, Speech → Text, Speaker Diarization. It accepts audio, video, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Scribe v3 is ElevenLabs' batch speech-to-text model, cutting word errors about 27% versus Scribe v2 across 102 languages, at $0.20 per audio hour.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: About $0.20 per hour of audio. Indicative; confirm on your ElevenLabs plan.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Upgrading to Scribe v3 means editing one string. It runs on the same POST /v1/speech-to-text endpoint as Scribe v2, existing parameters work unchanged, and you only swap model_id to scribe_v3. ElevenLabs shipped it in the week of 9 November 2026 for recorded, batch audio.
The reason to swap: on FLEURS across 102 languages, mean word error rate falls from 16.94% to 12.38%, about 27% fewer errors, and it improves on every language tested.
ElevenLabs computes WER per clip and then averages per language. Many published figures are corpus-level, pooling all words before dividing. The two methods give different numbers, so do not line this result up against another vendor's headline WER. Run your own audio instead.
Two parameters arrive with v3:
numbers_format chooses digits or words for spoken numbers, which saves a post-processing step for finance, logistics and anything with order numbers.secondary_language_codes hints at the other languages in mixed-language recordings, helping with code-switching speakers.Everything else from v2 is kept: diarization for up to 32 speakers (with more accurate labels than before), word- or character-level timestamps, keyterm prompting for up to 1,000 terms, entity detection, audio event tagging, transcript editing, webhooks and export to SRT, DOCX and PDF.
| Field | Value |
|---|---|
| Model ID | scribe_v3 |
| Mode | Batch (recorded audio and video) |
| Languages | 90+ |
| Speakers | Up to 32 with diarization |
| Keyterms | Up to 1,000 |
| Indicative price | $0.20 per audio hour |
| Predecessor | scribe_v2, still available |
The family has split three ways. This model is the general-purpose batch choice. Scribe v3 Live is the streaming counterpart for voice agents and captions, on a separate WebSocket at a slightly higher rate. Scribe v2 Medical is a clinical fine-tune of the older generation. For meeting notes, subtitles, lecture capture and call analytics, v3 is now the default; for clinical dictation, test the medical model against it rather than assuming either wins.
Pull a hundred recent files from production, already corrected by humans if you have them. Transcribe them with v2 and v3 using identical keyterms and settings, then compute WER yourself against those references. Look at speaker-label accuracy on calls with three or more people; that is where v3 claims a clear gain. Check that your downstream parsers handle numbers_format the way you expect before you turn it on.
Alternatives on the same task include Voxtral Mini Transcribe 2 and other speech-to-text models; see also speaker diarization models.
Compare the model on the work you need to complete.
Scribe v3 is listed for Automatic speech recognition, Speech → Text, Speaker Diarization. It accepts audio, video, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
inclusionAI
Ming-3.0-asr-tiny is inclusionAI's open speech-recognition model for Chinese and English with hotword hints, dialect coverage and context-aware ASR.
ElevenLabs
Scribe v2 Medical is ElevenLabs' HIPAA-eligible batch transcription model for clinical audio, with 35% fewer errors than Scribe v2 on medical speech.