Tasks and inputs
Scribe v3 Live is listed for Automatic speech recognition, Speech → Text. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
Scribe v3 Live is ElevenLabs' streaming transcription model with turn prediction, mid-sentence language switching and a 10,000-term keyterm library.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: About $0.22 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
Silence is a poor signal that someone has finished speaking. People pause mid-thought, and voice agents that commit on silence either cut users off or wait too long. Scribe v3 Live, ElevenLabs' real-time transcription model released in the week of 9 November 2026, adds turn prediction: with commit_strategy=turn_prediction, it commits a segment when it predicts the speaker is done, not merely when the audio goes quiet.
That single feature is why many teams will try it. The accuracy gains are a bonus.
The pricing page rounds the first figure to 43% fewer errors. Same comparison, different rounding; neither is an independent result.
Multilingual mode. Set multilang_mode=true and it transcribes mid-sentence language switches, writing each language in its own script. Hindi-English or Spanish-English callers stop producing garbled transliteration.
Keyterm library. Up to 10,000 terms, consulted only when a likely match is heard, on top of 50 standard keyterms. Product catalogues and staff names finally fit.
Realtime diarization. diarize=true adds draft speaker labels to partial transcripts, with a separate diarized version for each committed segment.
Also supported: automatic language detection, PCM (8 to 48 kHz) and mu-law audio, manual and VAD commit strategies, word timestamps, detection and redaction of 65 entity types, and background audio filtering.
At launch, the official SDKs did not support turn prediction, multilingual mode or the keyterm library. To use them you talk to the raw WebSocket at wss://api.elevenlabs.io/v1/speech-to-text/realtime, passing scribe_v3_live as model_id. The default commit strategy is still manual, so existing integrations behave exactly as before until you opt in.
About $0.22 per audio hour, a little above batch transcription (indicative). Real-time concurrency is capped by plan, from 6 streams on Free to 45 on Scale and Business, so size the plan to peak simultaneous calls.
Two neighbours are easy to confuse. Scribe v3 is the batch model for recorded files. Eleven Realtime is a full speech-to-speech agent with its own LLM and voice. This model only turns live speech into text; you bring the rest of the pipeline.
Record twenty live sessions with natural pauses, interruptions and at least one language switch. Replay them through v2 Realtime with silence commits and through v3 Live with turn prediction. Count premature commits and measure the delay between end of speech and commit. Compare transcripts for keyterm hits. Benchmark against MAI-Transcribe-2-Streaming and Gemini 3.5 Transcribe Live on the same recordings, or see all speech recognition models.
Compare the model on the work you need to complete.
Scribe v3 Live is listed for Automatic speech recognition, Speech → Text. It accepts audio 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.
Microsoft
MAI-Transcribe-2-Streaming transcribes live audio with intermediate and final segments through Microsoft Foundry’s speech services.
assemblyai
AssemblyAI Universal-3.6 Pro is a realtime streaming speech-to-text model covering 32 languages with code switching, at $0.45 per audio hour.
fish-audio
Fish Audio Transcribe 1 Pro is a multi-speaker speech-to-text model with inline speaker labels, emotion cues and word timestamps, for files up to 60 minutes.
xAI
Grok Voice Transcribe 2.0 transcribes recorded and streaming audio; the retired 1.0 API name now routes to this revision.
Gemini 3.5 Transcribe Live from Google: Speech transcription model for accurate audio-to-text and captioning workflows.
xAI
Grok Voice STT 1.0 from xAI: Grok Voice STT 1.0 is xAI's speech-to-text model. It supports transcription with word-level timestamps, optional speaker diarization, and multichannel audio.