ServiceNow releases open Apriel Voice 27B: speech, reasoning and tool use in one model
ServiceNow-AI · hf · 2026-11-02
ServiceNow AI open-sources Apriel Voice 27B Preview, a speech-language model built on a post-trained Gemma 4 26B-A4B checkpoint that unifies speech understanding, reasoning, and tool calling in a single backbone — no separate dialogue model needed. A two-stage procedure with online distillation yields 98.90 on Big Bench Audio, 81.13 on Speech-BFCL, and 65.51% on τ-Voice in thinking mode, with 1.91s mean latency. Demo and custom serving platform available.
More from Multimodal
- Parametric Studios demos LTX 2.5 video generation with VFX — EqualLettuce3014 · 2026-11-02
- AI-generated chubby knight sprint video wows Reddit with realistic lighting and motion — love1008 · 2026-11-02
- Turning a home into a walkable 3D digital twin with just an iPhone — imdigitalashish · 2026-11-02
- 22 projects showing builders using SAM 3.1 in unexpected ways via Meta API — Vjeux · 2026-11-02
- Giant man on a tiny bike: AI generation now makes the impossible trivial — HeyZoyaKhan · 2026-11-02
- Realtime audio-reactive LENIA artificial life visuals debut in AG Cook club set — zzznah · 2026-11-02