Liquid AI Releases Two Open Decision Models, d1-3B and d1-Omni-600M

Liquid AI has released two open-weight multimodal decision models, d1-3B and d1-Omni-600M, part of its d1 model family, according to aitimes.com. Rather than generating tokens sequentially, the models compute an input state in one pass and return an immediate decision score.

d1-3B carries 3.12 billion parameters and scored 48.57 on the Decision Index v0.2.1 benchmark, ranking first among models with fewer than 10 billion parameters and surpassing Decider 35B-A3B at 47.11. It also averaged 82.9 points across seven text-focused public benchmarks, ahead of Decider 4B at 81.1.

The smaller d1-Omni-600M has 587 million parameters and averaged 78.4 points, beating Decider 2B at 77.1. It recorded top performance on Civil Comments at 95.8 and PAWS-X at 79.5.

On input handling, d1-3B processes text and images with a context of 32,768 tokens, while d1-Omni-600M analyzes text, images and audio of up to 30 seconds in a single computation. d1-3B completes a decision in 8 milliseconds on an Nvidia RTX 4090 and around 50 milliseconds on an Nvidia Jetson Orin Nano.

Availability and licensing

Both models are downloadable from Hugging Face and can run locally or be fine-tuned for deployment. Under Liquid AI's LFM open license, companies with annual revenue under 10 million dollars can use them commercially for free. Listed use cases include AI model routing, harmful content detection, search result re-ranking, LLM-as-a-judge evaluation, AI agent guardrails, and real-time image and voice inspection. The models support decision modes named noul, choice and score.