ServiceNow releases open Apriel Voice 27B preview for voice-driven tasks
ServiceNow has released Apriel Voice 27B, an open-weight preview that takes text, images and voice input and can act on what it understands.
ServiceNow has published Apriel Voice 27B, an open-weight preview model on Hugging Face. The company describes it as built to get work done from voice input. Berman, who covered the release in a video, said the model accepts text, images and voice together and combines them. He said it then decides what to do based on what it understands the user is trying to accomplish. He also said it can produce text-to-speech output.
The architecture, as Berman walked through it, has several parts. Speech enters through a speech encoder, text through token embeddings, and images through what he called a frozen visual interface. All of these feed into a mixture-of-experts backbone. On performance, Berman said the model scores very well on benchmarks for its size, though the video gave no specific figures.
What changed
ServiceNow released Apriel Voice 27B, an open-weight preview model that takes text, image and voice input.
Type: New capability · Applies to: Apriel Voice 27B
Why it matters
An open voice-first agent model could let enterprises build voice workflows on infrastructure they control.
What you can do
Teams exploring voice agents could assess whether the preview is ready for internal evaluation.
Original source
We have no link to the original source for this story yet.
Where we saw it
Forward Future (Matthew Berman)
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works