← Models

Scribe v2

ElevenLabs · Scribe

ElevenLabs' current speech-to-text model line for accurate batch transcription and low-latency realtime STT.

Type
audio
Context
N/A
Max Output
N/A
Status
current
API Access
Yes
License
proprietary
speech-to-text transcription audio diarization timestamps realtime multilingual elevenlabs
Released January 2026 · Updated June 8, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on June 8, 2026.

Scribe v2 is ElevenLabs’ current speech-to-text model line. It belongs on Signal Lens as a model entry because ElevenLabs now documents stable model IDs, pricing, and distinct batch versus realtime behavior rather than treating transcription as a generic platform feature.

There are two practical routes: scribe_v2 for accurate asynchronous transcription, subtitling, and captioning, and scribe_v2_realtime for live transcription in voice agents, meetings, captions, and conversational systems. This page covers the Scribe v2 generation as a whole, with the route distinction called out where it affects architecture and cost.

Capabilities

Scribe v2 is built for real production audio rather than clean demo clips. ElevenLabs documents support for 90+ languages, keyterm prompting, entity detection, precise word-level timestamps, speaker diarization, dynamic audio tagging, and smart language detection.

That makes it useful when transcripts need to feed downstream workflows: editing, subtitle alignment, compliance review, searchable archives, agent memory, or structured extraction. Scribe v2 Realtime trades some of the batch-oriented posture for latency, with ElevenLabs positioning it around roughly 150 ms live transcription and use in agents, meeting notetakers, and live applications.

Technical Details

For speech-to-text models, token context and max-output numbers are not meaningful in the same way they are for language models. Signal Lens stores contextWindow: 0 and maxOutput: 0 intentionally for this page.

Current official model IDs:

  • scribe_v2 for state-of-the-art batch speech recognition
  • scribe_v2_realtime for realtime speech recognition

Current docs list:

  • 90+ supported languages
  • Keyterm prompting up to 1000 terms
  • Entity detection up to 56 categories
  • Speaker diarization up to 32 speakers
  • Precise word-level timestamps
  • Dynamic audio tagging and smart language detection

Pricing & Access

Current ElevenLabs API pricing lists:

  • Scribe v1/v2 speech-to-text: $0.22 per hour
  • Entity detection add-on: $0.070 per hour
  • Keyterm prompting add-on: $0.050 per hour
  • Scribe v2 Realtime: $0.39 per hour

Scribe v2 is available through the ElevenLabs API and Creative platform. Realtime access is designed for streaming STT flows and should be budgeted separately from any LLM, voice-agent, or text-to-speech costs in the full application.

Best Use Cases

Use Scribe v2 for podcast and video transcription, captions, research interviews, support-call processing, compliance review, media archives, and multilingual content operations. Use Scribe v2 Realtime when the transcript has to drive a live agent, live caption, meeting assistant, or translation-like workflow.

The main caution is governance. If transcripts include health data, payment information, personal identifiers, or confidential meetings, treat retention, redaction, access control, and review policy as first-class parts of the pipeline.

Comparisons

  • OpenAI GPT-4o Transcribe: Strong fit when the broader workflow already sits in OpenAI’s realtime or transcription stack.
  • Deepgram Nova-style STT: Often attractive for low-latency transcription pipelines with speech-specialist controls.
  • ElevenLabs voice stack: Scribe v2 is strongest when paired with ElevenLabs text-to-speech, dubbing, or voice-agent workflows.