# Liquid AI releases Apple‑Silicon version of d1‑3B decision model

> A float16 export of the d1‑3B model is now on Hugging Face, optimized for Apple GPUs and usable via the mlx‑vlm library.

Oossa · 2026-10-12 · https://oossa.com/en/liquid-ai-releases-apple-silicon-version-of-d1-3b-decision-model

Liquid AI has posted a new version of its d1‑3B decision model on Hugging Face. The model, called LiquidAI/d1-3B-MLX-fp16, runs in float16 precision and is built for Apple Silicon devices. It can answer multiple typed questions about text, JSON or images in a single forward pass, without generating tokens. The repository includes install instructions for the mlx‑vlm library and examples for both text and image inputs.

## The facts

- Model size: 6.25 GB
- Published: 2026‑10‑12 (Mon Oct 12 2026 03:55:52 GMT+0200)

## Why it matters

Developers can now run complex decision‑making AI locally on Macs, avoiding cloud costs and latency.

## Sources & references

1. [New model on Hugging Face: LiquidAI/d1-3B-MLX-fp16 (image-text-to-text)](https://huggingface.co/LiquidAI/d1-3B-MLX-fp16) – Liquid AI, 2026-10-12
2. [New model on Hugging Face: LiquidAI/d1-3B-MLX-4bit (image-text-to-text)](https://huggingface.co/LiquidAI/d1-3B-MLX-4bit) – Liquid AI, 2026-10-12
3. [New model on Hugging Face: LiquidAI/d1-3B-MLX-8bit (image-text-to-text)](https://huggingface.co/LiquidAI/d1-3B-MLX-8bit) – Liquid AI, 2026-10-12

Last updated: 2026-10-12