Many MindedElevenLabs adds live transcription that reads context to know when you're done
ElevenLabs has released two speech-to-text models: Scribe v3 for pre-recorded audio and Scribe v3 Live for low-latency, real-time use. Both are available through the ElevenAPI developer interface and the ElevenAgents platform.
The company says Scribe v3 cuts word errors by about 25 percent against its predecessor across 102 languages, with notably better German accuracy, and that speaker diarization is more reliable in larger group conversations. Scribe v3 Live offers diarization in real time for the first time. It also analyzes conversational context along with the audio to judge when a speaker has finished, rather than relying on rigid pause timers. Developers can preload thousands of specialized terms such as brand names or medications, set natural-language formatting rules for numbers and dates, and use a multilingual mode that handles mid-sentence language switches. Enterprise customers get isolated EU servers and a zero-retention mode that deletes data right after each request.
Why it matters: voice agents that interrupt callers or stumble over accents, medications and language switches are hard to use, and the people most affected are often those who rely on voice rather than screens. Better turn-taking and recognition make voice services more usable and widen who can reach them. ElevenLabs is also running a phone agent for Poland's public health system under an expanded Nvidia partnership.
Caveats: the accuracy, diarization and German gains are ElevenLabs' own claims. Artificial Analysis ranks Scribe v3 ahead of models from Microsoft, Mistral, xAI, Google and OpenAI, behind only an Alibaba preview model, but that benchmark covers only word errors on recorded files, not diarization, rare languages or Scribe v3 Live's real-time behavior. At $3.33 per 1,000 minutes, Scribe v3 costs roughly twice what faster Microsoft and xAI models do, so it sits mid-pack on speed and price.
Source: The Decoder