ServiceNow previews Apriel Voice 27B speech-reasoning model
AI summary
ServiceNow announced Apriel Voice 27B Preview, an open-weight model built around a Gemma 4 26B-A4B backbone with a Whisper large-v3-turbo speech encoder. It accepts audio, text, and images, and is designed to produce text responses and structured tool calls when paired with an external speech synthesizer and application-managed turn-taking. The model card also reports speech-recognition, reasoning, tool-use, and latency benchmarks.
In-depth insight
Apriel Voice 27B Preview combines speech understanding, reasoning, multimodal context, and structured tool calling in an open-weight release, offering developers a model for voice assistants, customer-support workflows, and spoken tool use.