Liquid AI releases open weight multimodal decision models d1 3B and d1 omni 600M

Liquid AI released two open-weight multimodal models in its d1 decision model family on October 7, 2026, MarkTechPost reported. The releases include d1-3B, which processes text and images, and d1-omni-600M, which reads text combined with either image or audio inputs. MarkTechPost noted that neither model generates output text, instead returning calibrated, typed answers in a single forward pass with zero output tokens. [1]

Liquid AI was founded in 2023 as an MIT spin-off by Ramin Hasani, Mathias Lechner, and Alexander Amini, alongside MIT computer scientist Daniela Rus. The American artificial intelligence company specializes in developing liquid foundation models designed to run directly on edge devices without requiring a cloud connection. [3]

Unlike standard large language models that generate, summarize, and translate text, the new d1 architecture bypasses token generation entirely. Conventional transformer architectures convert input data such as text, images, or audio into tokens and process them using parallel multi-head attention mechanisms to create vector representations. [1][2][4]