ElevenLabs releases Scribe v3 and real-time Scribe v3 Live
Original title: ElevenLabs Scribe v3 Live uses conversational context to stop voice agents from cutting off callers mid-sentence
The company claims about 25 percent fewer word errors across 102 languages; independent tests rank accuracy high but speed mid-pack.
ElevenLabs released Scribe v3 for recorded audio and Scribe v3 Live for low-latency use, both available through ElevenAPI and ElevenAgents. The models target overlapping speech, noise, speaker diarization and voice agents that previously cut callers off on short pauses. Artificial Analysis ranks Scribe v3 among the most accurate transcription models, behind only an Alibaba preview, while Microsoft and xAI are faster and cheaper. ElevenLabs is also expanding enterprise work with Nvidia, McKinsey and a voice-agent marketplace.
Unconfirmed: The roughly 25 percent error cut, German accuracy and diarization gains are company claims; the cited benchmark covers recorded-file word errors only.
- Scribe v3 Live uses conversational context to judge when a speaker has finished.
- Artificial Analysis lists a benchmark price of $3.33 per 1,000 minutes.
- Developers can preload specialized terms and set number or date formatting in natural language.
- Enterprise options include EU isolated servers and a zero-retention mode.
- Nvidia work includes Nemotron 3.5 Super VL in ElevenAgents and a Poland health phone agent.
- Vendors
- ElevenLabsNvidia
- Products
- ElevenAPIElevenAgents
- Organizations
- Artificial AnalysisMcKinsey
- Countries
- Poland
From the original article
2 min read at the-decoder.com
ElevenLabs has released Scribe v3 and Scribe v3 Live, two new speech recognition models designed to transcribe accurately through background noise and crosstalk. Benchmarks from Artificial Analysis confirm that Scribe v3 narrowly beats competitors like OpenAI, Google, and Microsoft on accuracy. That precision comes at a cost, though.
By Maximilian Schreiner · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.