LiquidAI d1 brings small open decision models to edge deployment

LiquidAI's d1-3B and d1-omni-600M are open-weight, non-generative models for routing, classification, moderation, and agent control, with reported llama.cpp and standard serving-framework support. Vendor-reported measurements span RTX 4090, Jetson, Apple, and AMD hardware, but independent benchmarks, full methodology, licensing confirmation, and the omitted private vision evaluation remain pending.