ElevenLabs launches Scribe v3 and Scribe v3 Live, its most accurate speech-to-text models yet

📅 November 12, 2026 • AI Automation
November 12, 2026 — On November 10, ElevenLabs launched two new speech-to-text models, Scribe v3 for recorded audio and Scribe v3 Live for realtime transcription, and developer and press attention to the release has continued through this week. Scribe v3 targets recorded material such as meeting recordings, interviews and call archives, while Scribe v3 Live handles live calls and voice agents with low latency.
Both models are available through ElevenAPI, Scribe v3 Live also works inside ElevenAgents, and switching from the previous generation is a one-line change of the model ID to scribe_v3 or scribe_v3_live. Scribe v3 is now the default transcription model in the ElevenLabs app.
Accuracy is the core of the launch. Across 102 languages, Scribe v3 makes 27 percent fewer word errors than Scribe v2, and on recordings with three or more speakers it also makes 17 percent fewer diarization errors, which means fewer mix-ups about who said what in multi-speaker transcripts. Scribe v3 Live makes 43 percent fewer errors than Scribe v2 Realtime, and it brings speaker diarization to realtime transcription for the first time, with 39 percent fewer diarization errors than the next-best realtime model ElevenLabs tested.
Turn detection in the Live model listens to the conversation itself rather than only to silence, so a caller who pauses mid-sentence is not cut off.
The models also add control features aimed at production pipelines. Developers can choose whether numbers are returned as digits or words, attach a keyterm prompt of up to 1000 terms, define transcript edits through formatting rules written in plain English, and switch on a multilingual mode for mid-sentence language changes. On the deployment side, processing runs in the US by default, while isolated environments in the EU, India and Singapore serve enterprise data residency requirements.
Zero Retention Mode deletes audio and text once a request completes, the stack carries SOC 2, ISO 27001 and PCI DSS Level 1 certifications, it is HIPAA-eligible with a BAA, and a version of Scribe v3 Live optimized for medical conversations is available in ElevenAgents.
Independent checks put the claims in context. On the Artificial Analysis board for recorded audio, read on November 12, Scribe v3 ranked second among 64 models at 1.9 percent word error, behind Alibaba Cloud Fun-Realtime-ASR-preview at 1.7 percent. Pricing is unchanged versus the previous generation: 0.20 dollars per hour for Scribe v3 and 0.22 dollars per hour for Scribe v3 Live.
For teams building voice automation, transcripts are the first stage of the pipeline that feeds meeting notes, contact centers, clinical documentation and voice agents, so fewer errors at the transcription stage mean fewer errors everywhere downstream.
Sources: elevenlabs.io — ai.info — cellcog.ai

Continue the conversation: https://t.me/AIAutomationArchitecture · Bluesky