Liquid AI releases open-weight d1 decision models with vision and audio support

What is changing?

Liquid AI has released d1-3B and the experimental d1-omni-600M from its d1 family as open-weight models. Unlike generative models, d1 does not produce output tokens; it returns structured decisions for tasks such as classification, routing, and scoring in a single forward pass. d1-3B accepts text and images, while the 600-million-parameter d1-omni handles text with either images or audio.

In the company's tests, d1-3B scored 48.57 on Decision Index 0.2.1. It answered a single question in 16 ms on Jetson AGX Thor, 26 ms on AGX Orin, and 50 ms on Orin Nano. Liquid AI explicitly notes that the release does not report vision or audio decision benchmarks and that d1-omni-600M remains an early research release.

Both models are downloadable from Hugging Face, and Liquid AI has collected ten camera, drawing, and text demos in its System One Arcade. The design could move narrow tasks such as content moderation, request routing, and real-time visual checks onto edge devices without a constant cloud connection.