What changed
ElevenLabs has released Scribe v3 and Scribe v3 Live, two speech recognition models built for messy audio situations like crosstalk and background noise, covering both pre-recorded files and real-time use. The real-time version analyzes conversational context to avoid cutting callers off during brief pauses. Both models are available now through the ElevenAPI developer interface and the ElevenAgents platform.
What this means in practice
Scribe v3 is designed for pre-recorded audio files, while Scribe v3 Live targets low-latency real-time applications. According to Elevenlabs, Scribe v3 cuts word errors by about 25 percent compared to its predecessor across 102 languages, with significant improvements in German accuracy. Speaker diarization also works more reliably in larger group conversations, and Scribe v3 Live makes this available in real time for the first time. Developers can preload the models with specialized terms like brand names or medications to avoid mishearing domain-specific vocabulary. They can also set formatting rules for numbers or dates in natural language, and a multilingual mode handles language switches mid-sentence. Enterprise customers get isolated servers in the EU and a zero-retention mode designed to delete data immediately after each request.
Limitations and availability
Independent benchmarks confirm that Scribe v3 is among the most accurate transcription models on the market, but it runs slower and costs more than rival models from Microsoft and xAI. The benchmark measures only word errors on recorded files and does not account for improvements in speaker diarization, rare languages, or the real-time behavior of Scribe v3 Live. Elevenlabs is also expanding its enterprise business alongside the new models, including partnerships with Nvidia and McKinsey, and a broader partner program for system integrators.
