Liquid AI Releases Open d1 Models for Edge Decisions
Liquid AI has released open-weight multimodal decision models designed to make fast structured choices on phones, PCs and edge hardware.
What happened
Liquid AI released two open-weight decision models, d1-3B and d1-omni-600M, through Hugging Face. The models are designed to output structured decisions rather than long natural-language explanations, accepting text and, in the larger model’s case, visual inputs. The company positions them for classification, routing and other situations where an AI system must choose an action quickly.
In testing conducted with NVIDIA, d1-3B answered a single question in under 50 milliseconds on the measured edge devices, including the Jetson AGX Thor, Jetson AGX Orin and Orin Nano. On an RTX 4090, Liquid AI reported response times below 10 milliseconds. The model can also process multiple questions with limited additional latency, a property that could matter for high-frequency control loops.
Liquid AI says d1-3B delivers the strongest decision quality in the release while d1-omni-600M is intended for environments where memory and compute are more constrained. Both models are downloadable from Hugging Face and require custom model code when loaded through Transformers.
Why it matters
Decision models are a narrower direction than general chat models, but that narrowness is precisely their appeal. A system that returns a calibrated choice or action can be cheaper and easier to validate than a model that generates open-ended text and then requires another component to interpret it. The unresolved question is generality: the reported latency and benchmark results show attractive edge behavior, but they do not establish that d1 models can safely handle ambiguous, high-stakes decisions outside the tested tasks. Their importance will be determined by whether developers use them as dependable control components rather than miniature chatbots.