Segment Anything (SAM)

weights

open(SAM 3 and SAM 3.1 checkpoints on the Hub behind a manual access request)

data

documented-not-released(the SA-Co training data from Meta's data engine is described

code

partial(inference and fine-tuning code

license

SAM-License(code and weights

SAM 3 and SAM 3.1 come under Meta's SAM License, which permits commercial use but bars military, weapons and trade-controlled uses and lets Meta change the terms, and the checkpoints need an approved access request. Meta publishes fine-tuning code and evaluation benchmarks but not the training data or pretraining code. SAM and SAM 2 were Apache-2.0, but the current release governs. The SAM License caps neither who may use the model nor at what scale, so its tier is permissive_non_osi.

• https://huggingface.co/api/models/facebook/sam3 recorded 2026-09-27

  Hub record for facebook/sam3, gated "manual", license other.

• https://huggingface.co/api/models/facebook/sam3.1 recorded 2026-09-27

  Hub record for facebook/sam3.1, gated "manual", license other.

• https://raw.githubusercontent.com/facebookresearch/sam2/main/README.md recorded 2026-09-27

  SAM 2 README: "The SAM 2 model checkpoints, SAM 2 demo code ... and SAM 2 training code are licensed under Apache 2.0" - the superseded release.

• https://raw.githubusercontent.com/facebookresearch/sam3/main/LICENSE recorded 2026-09-27

  LICENSE is the "SAM License", "Last Updated: November 19, 2025"; 1.a grants "a non-exclusive, worldwide, non-transferable and royalty-free limited license ... to use, reproduce, distribute, copy, create derivative works of, and make modifications to the SAM Materials"; 1.b bars trade-controlled, military, nuclear, espionage and weapons uses. No non-commercial clause.

• https://raw.githubusercontent.com/facebookresearch/sam3/main/README_TRAIN.md recorded 2026-09-27

  README_TRAIN: "This repository supports finetuning SAM3 models on custom datasets in multi-node setup or local execution."

• https://raw.githubusercontent.com/facebookresearch/sam3/main/README.md recorded 2026-09-27

  README: "This project is licensed under the SAM License"; "please request access to the checkpoints on the SAM 3 Hugging Face repo"; "03/27/2026 -- SAM 3.1 Object Multiplex is released"; "We release 2 image benchmarks, SA-Co/Gold and SA-Co/Silver, and a video benchmark SA-Co/VEval"; the data engine "automatically annotated over 4 million unique concepts".

Hugging Face downloads over the trailing 30 days for the two SAM 3 releases, which are the ones declared. The superseded SAM 2 and original SAM checkpoints add more and are not declared. The sam3 PyPI package, which installs the code, is declared but not counted.

• https://huggingface.co/api/models?author=facebook&search=sam&sort=downloads&limit=100&expand[]=downloads&expand[]=cardData&expand[]=gated recorded 2026-09-27

  facebook/sam3 2,105,970 and facebook/sam3.1 49,010 downloads in the trailing 30 days (2,154,980 together); sam-vit-huge 328,473, sam-vit-base 240,441 and the sam2 and sam2.1 checkpoints listed alongside under apache-2.0.

• https://pypistats.org/api/packages/sam3/recent recorded 2026-09-27

  last_month 71,753 downloads of sam3, the code package, not added to the Hub figure.

SAM 3 segments whatever the user names or points at in their own images and video with no training, the same no-training capability that places Ultralytics here. Ultralytics sits here partly because it bundles SAM 3; the model line itself does one family of tasks, segmentation and tracking, where Ultralytics spans many.

• https://raw.githubusercontent.com/facebookresearch/sam3/main/README.md recorded 2026-09-27

  README: "SAM 3 is a unified foundation model for promptable segmentation in images and videos. It can detect, segment, and track objects using text or visual prompts such as points, boxes, and masks." and "SAM 3 introduces the ability to exhaustively segment all instances of an open-vocabulary concept specified by a short text phrase or exemplars".