LiquidAI releases open d1 decision models for edge inference on Hugging Face

LiquidAI today released two open decision models—d1-3B and d1-omni-600M—as part of its d1 decision model family on Hugging Face. Unlike generative models, these decision models produce answers in a single forward pass rather than generating tokens sequentially.

d1-3B achieves a Decision Index 0.2.1 score of 48.57, surpassing Decider 35B-A3B (47.11) and every 4B and 9B model listed. It supports text and image inputs and was trained from the LiquidAI LFM2.5-VL-3B backbone, a decoder-only vision-language model.

d1-omni-600M, currently in an early research release, supports either text+image or text+audio inputs and was trained from the LFM2.5-Encoder-350M bidirectional encoder with added vision and audio encoders. It scores 78.4 on the benchmark mean, outperforming Decider 2B (77.1) with roughly a quarter of the parameters.

On edge hardware, d1-3B answers a single question in 16 ms on NVIDIA Jetson AGX Thor, 26 ms on Jetson AGX Orin, and 50 ms on Jetson Orin Nano. On GPU, it answers in under 10 ms and processes a 384px image in under 18 ms across tested platforms.

The models were benchmarked on seven public datasets spanning reading comprehension, toxicity detection, intent classification, medical QA, and cross-lingual understanding. d1-3B’s mean score across these datasets is 82.9, the highest in the reported table and above Decider 4B.

Both models are open-weight and available for download on Hugging Face. LiquidAI provides demos in the System One Arcade Hugging Face Space, and usage examples demonstrate single-pass structured decision-making with multimodal inputs and batch processing.