Intel / article

Multimodal open d1 decision models for the edge

Source: huggingface.co
Date: Published Oct 7, 2026

From the Daily Brief

Decision models return a typed answer in one forward pass instead of generating tokens, which makes them suited to routing, reranking, guardrails and visual inspection inside agent systems. Liquid AI's d1-3B handles text and vision, while the experimental d1-omni-600M accepts text with images or audio. Liquid reports latency from about 8 ms on an RTX 4090 to under 50 ms on a Jetson Orin Nano, along with benchmark gains over existing Decider models. Open weights and llama.cpp support on day one mean these classifiers can run locally rather than as a hosted API call. The release adds to a crowded week for small decision models, with Strands Decider 2B released and Cloudflare's Clef landing on OpenRouter, so teams choosing one should test on their own routing labels.

Read the 2026-10-08 Brief →

Context

Liquid AI released two models in its d1 family, described in its Hugging Face post. Unlike a generative model that writes text, a d1 decision model answers in a single forward pass, one run through the network, with a structured decision.

d1-3B takes text and images. The experimental d1-omni-600M takes text with an image or with audio and is an early research release. Liquid reports that d1-3B scores 48.57 on its Decision Index 0.2.1, ahead of the larger models it lists, averages 82.9 across seven public datasets, and answers a question in 16 milliseconds on a Jetson AGX Thor edge computer and 8 milliseconds on an RTX 4090. These are Liquid's own numbers.

Liquid recommends d1 models when an application needs fast, structured decisions, including from inputs.

Terms in this piece · Glossary

open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Read the source huggingface.co