AI News
Liquid AI releases open d1 decision models for edge inference
Liquid AI released d1-3B and d1-omni-600M, open decision models that answer in a single forward pass. Benchmarks and edge speed details from the Hugging Face blog.
What this piece is grounded in
01
Today, we release two open decision models in our d1 decision model family : d1-3B and d1-omni-600M (experimental).
02
d1-3B achieves a mean score of 82.9, the highest in the table and above Decider 4B.
03
d1-omni-600M scores 78.4, surpassing Decider 2B (77.1) with only a quarter of the parameters.
04
d1-3B answers a single question in under 50 ms on every measured device.
01 / FIELD NOTE
What changed
Liquid AI released two open decision models in its d1 family: d1-3B and d1-omni-600M (experimental), according to a Hugging Face blog post published on October 7, 2026. The post states that unlike generative models, these decision models do not produce tokens but answer in a single forward pass. The models are built on Liquid Foundation Models (LFMs) and are open-weight, available on Hugging Face. The release includes benchmark results on seven public datasets covering reading comprehension, toxicity detection, intent classification, medical QA, and cross-lingual understanding.
02 / FIELD NOTE
Why it matters for practical AI work
The reported benchmark results suggest these models are designed for fast, structured decisions, including multimodal inputs. The blog reports d1-3B achieves a mean score of 82.9, the highest in the table and above Decider 4B. d1-omni-600M scores 78.4, surpassing Decider 2B (77.1) with only a quarter of the parameters. For edge inference, d1-3B answers a single question in under 50 ms on every measured device, with three questions taking only 1.3x the time of one. On GPU, d1-3B answers a question in under 10 ms and processes a 384px image in under 18 ms on both platforms.
03 / FIELD NOTE
What a reader can verify
A reader can verify the model availability and benchmark details by visiting the Hugging Face blog post. The post includes instructions for using the models, noting that they require transformers>=5.14 and should be loaded with trust_remote_code=True. The blog also mentions that d1-3B was evaluated in collaboration with NVIDIA across NVIDIA GeForce RTX 4090, NVIDIA Jetson AGX Thor, Jetson AGX Orin 64 GB, and Jetson Orin Nano. The post states that d1-3B retains the vision capabilities of its LFM2.5-VL-3B backbone on standard vision benchmarks, and d1-omni-600M handles all three modalities.
04 / FIELD NOTE
What remains unclear
The supplied feed entry is a concise summary; consult the canonical article for full context. The blog does not report vision or audio benchmarks, as the Decision Index v0.3 includes only a private vision split and audio decision benchmarks are currently an open problem. Since d1-omni-600M is an early research release, no speed numbers are reported for it. The post does not provide details on the training data, the exact architecture differences between the two backbones, or the full model card instructions for d1-omni-600M, which are referenced but not included in the summary.
Questions readers ask
What are d1 decision models?
According to the Hugging Face blog, d1 decision models are part of Liquid AI's d1 family. Unlike generative models, they do not produce tokens but answer in a single forward pass. They are built on Liquid Foundation Models and are designed for fast, structured decisions, including multimodal inputs. Two models were released: d1-3B and d1-omni-600M (experimental).
How fast is d1-3B on edge devices?
The blog reports that d1-3B answers a single question in under 50 ms on every measured device, including NVIDIA GeForce RTX 4090, Jetson AGX Thor, Jetson AGX Orin 64 GB, and Jetson Orin Nano. Three questions take only 1.3x the time of one, with the AGX Thor going from 16 ms to 20 ms. On GPU, it answers a question in under 10 ms.