Liquid AI releases compact d1 decision models for edge inference
Liquid AI released weights for d1-3B and experimental d1-omni-600M, compact models that return structured decisions rather than generated prose. The 3B model handles text and images; the 600M variant adds audio, with vendor benchmarks and edge-latency measurements published alongside the code path.
Why it matters: Single-pass classifiers can be faster and easier to calibrate than generative prompts for routing, ranking and policy decisions, while multimodal inputs make the pattern relevant to local devices and real-time agents.