facebook · First MLX build 24 Dec 2025 · In Core AI since 8 Jun 2026

sam3

MLXCore AI Vision other

A 860M parameter segmentation model from facebook, with 9 MLX builds, from 4-bit to bf16 and Apple Core AI export recipes for iOS, macOS.

Show details for
Hugging Face Apple recipe
Parameters
860M
From Hugging Face
Quantisation
8-bit
sam3-8bit
Memory needed
4 GB
Estimate, unified memory
Download
985 MB
Weights, estimated
Downloads
208
This build, last 30 days
Task
Segmentation
Vision
MLX builds
9
On mlx-community
Devices
Mac
Apple silicon

Memory

Estimated from the size of the weights, plus a fifth for context and working memory and 2 GB for macOS and your apps. Long contexts need more.

Run it

Terminal · Hugging Face CLI
$ pip install -U huggingface_hub
$ hf download mlx-community/sam3-8bit

MLX builds

9 on mlx-community
Click a column to sort. Smallest size and most downloads are highlighted.
sam3-4bit 4-bit 568 MB 3 GB 85
sam3-mxfp4 MXFP4 568 MB 3 GB 24
sam3-nvfp4 4-bit 568 MB 3 GB 22
sam3-5bit 5-bit 672 MB 3 GB 32
sam3-6bit 6-bit 777 MB 3 GB 27
sam3-8bit 8-bit 985 MB 4 GB 208
sam3-mxfp8 MXFP8 985 MB 4 GB 34
sam3-bf16 bf16 1.7 GB 4 GB 152
sam3-image Unstated 3.4 GB 6 GB 0

Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.