ElevenLabs has introduced two new speech-to-text models. Scribe v3 is intended for pre-recorded audio files, while Scribe v3 Live is built for low-latency, real-time applications. Both are designed for difficult audio conditions, such as overlapping speakers in meetings or phone calls with background noise, and both are available now through the ElevenAPI developer interface and the ElevenAgents platform.
According to the company, Scribe v3 reduces word errors by roughly 25 percent compared with its predecessor across 102 languages, with German accuracy in particular said to improve significantly. Speaker diarization is also said to work more reliably in conversations with larger groups, and Scribe v3 Live brings diarization to real-time transcription for the first time.
Benchmark results
Independent testing by Artificial Analysis places Scribe v3 among the most accurate transcription models on the market. It scores ahead of competing models from Microsoft, Mistral, xAI, Google and OpenAI, with only a preview model from Alibaba ranking higher.
The same benchmark shows trade-offs, however. At $3.33 per 1,000 minutes, ElevenLabs sits in the middle of the pack on both speed and price: models from Microsoft and xAI run many times faster, produce only marginally higher error rates and cost about half as much. Anyone looking to process large volumes of audio quickly and cheaply may therefore be better served by other providers.
The benchmark also measures only word errors on recorded files. It does not account for the improvements ElevenLabs emphasizes in speaker diarization, rare languages, or the real-time behavior of Scribe v3 Live.
A real-time model for voice agents
For voice agents, Scribe v3 Live is meant to address a known weakness: current systems often rely on rigid pause timers and can cut callers off mid-sentence when they pause briefly to think or slowly read out a string of numbers. The new model analyzes conversational context together with the audio signal to determine whether a speaker has actually finished.
ElevenLabs also gives developers more control over the final transcript. The models can be preloaded with thousands of specialized terms, such as brand names or medications, to avoid mishearing domain-specific vocabulary. Formatting rules for numbers or dates can be set 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.
Enterprise expansion
Alongside the new models, ElevenLabs is expanding its enterprise business. Under an expanded partnership with Nvidia, the company is integrating the open Nemotron 3.5 Super VL language model into ElevenAgents, securing additional compute capacity running on Nvidia, developing local voice control for robots, and running a phone agent for Poland's public health system. The company also announced an alliance with consulting firm McKinsey, a broader partner program for system integrators, and its own marketplace where companies like Salesforce, IBM and Zendesk offer ready-made voice agents.