Neuphonic ships NeuDecide: 43MB audio-first decision model, 46ms inference on CPU, offline
tobowers · x · 2026-10-09
Neuphonic released NeuDecide, an audio-first tiny decision model at just 43MB that runs on a single CPU thread. It maps speech directly to tool calls (no transcription needed), achieves 46ms inference on an M3 MacBook, runs fully offline on mobile and even Raspberry Pi, and scores 75.5% tool accuracy on SLURP.
Related event: Neuphonic open-sources 43MB NeuDecide for direct speech-to-tool-calls(2 posts)→
More from Models
- Developer says he now manages 90% of his coding work through Grok bot — jonathan_wilke · 2026-10-09
- State of AI Report 2026 Predicts AGI by 2027 and Lists 9 Bets for the Next Year — FinanceYF5 · 2026-10-09
- Anthropic Leads on Coding but Not Overall: Astra Still Ahead, Astra 6.1 Coming — haider1 · 2026-10-09
- ChatGPT's $500 tier measured: weekly cap hits at ~90M Sol + 25M Luna tokens — airesearch12 · 2026-10-09
- Opus Now Imitates Your Voice in Video Easier Than Writing Human-Sounding Prose — tobowers · 2026-10-09
- Jev works surprisingly well as a cheap general-purpose classifier, dev reports — max_paperclips · 2026-10-09