Liquid AI has released two open-weight models under the name Open d1, part of its d1 decision model family. The larger d1-3B takes text and image inputs, while d1-omni-600M accepts text with an image or text with audio. Both are designed for decision-making tasks rather than open-ended generation.
What sets these models apart is that they do not write text at all. Instead of producing a sequence of output tokens, each model returns calibrated, typed answers in a single forward pass. That design could reduce latency and compute cost for applications that only need a structured result, such as classification or scoring.
The weights are open, so developers can inspect, run, and adapt the models. The two variants cover different input modalities, but they share the same zero-output-token approach, making the family consistent in behavior while differing in scale and sensor inputs.