Depth Anything

weights

open(every DA3 checkpoint on the Hub, ungated)

data

documented-not-released(stated to be public academic datasets only

code

partial(inference, CLI and benchmark code

license

Apache-2.0(code, and the Small, Base, Metric and Mono checkpoints)+CC-BY-NC-4.0(the Large, Giant and Nested any-view checkpoints)

The code and the small, base, metric and monocular models are Apache-2.0, but the large, giant and nested any-view models, the line's flagships, are CC BY-NC 4.0 and may not be used commercially. The map treats the most restrictive license among a release's distributed checkpoints as the release's. One Hub card, DA3-LARGE-1.1, is tagged Apache-2.0 while the repository's license table lists it as CC BY-NC; the giant models settle the reading either way.

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

  All depth-anything DA3 checkpoints are gated false; DA3-GIANT, DA3-LARGE and DA3NESTED-GIANT-LARGE-1.1 tagged cc-by-nc-4.0, DA3METRIC-LARGE, DA3MONO-LARGE, DA3-BASE and DA3-SMALL apache-2.0.

• https://huggingface.co/depth-anything/DA3-GIANT-1.1/raw/main/README.md recorded 2026-09-27

  DA3-GIANT-1.1 card: license cc-by-nc-4.0 and "Non-commercial use only".

• https://huggingface.co/depth-anything/DA3-LARGE-1.1/raw/main/README.md recorded 2026-09-27

  DA3-LARGE-1.1 card: "license: apache-2.0", contradicting the repository table's CC BY-NC 4.0 for that checkpoint.

• https://raw.githubusercontent.com/ByteDance-Seed/Depth-Anything-3/main/LICENSE recorded 2026-09-27

  LICENSE is the Apache License, Version 2.0, "Copyright 2025 The Depth Anything 3 Team".

• https://raw.githubusercontent.com/ByteDance-Seed/Depth-Anything-3/main/README.md recorded 2026-09-27

  README model table: DA3NESTED-GIANT-LARGE(-1.1), DA3-GIANT(-1.1) and DA3-LARGE(-1.1) "CC BY-NC 4.0"; DA3METRIC-LARGE "Apache 2.0"; "All models are trained exclusively on public academic datasets."

• https://ungh.cc/repos/ByteDance-Seed/Depth-Anything-3/files/main recorded 2026-09-27

  Repository tree of 133 paths holding inference, CLI, benchmark, app and streaming code; no training script.

Hugging Face downloads over the trailing 30 days for the six declared Depth Anything 3 checkpoints. The older V2 and V1 checkpoints, Depth-Anything-V2-Small above all, are downloaded more than the current release; they would raise the total without changing its order of magnitude.

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

  DA3METRIC-LARGE 259,633, DA3-LARGE-1.1 258,562, DA3MONO-LARGE 207,983, DA3NESTED-GIANT-LARGE-1.1 129,398, DA3-LARGE 103,007 and DA3-GIANT 100,195 downloads in the trailing 30 days (1,058,778 together); Depth-Anything-V2-Small-hf 2,711,667.

Depth Anything gives a depth map, and with several views a camera pose, for any image the user supplies without training, and depth has no label set that a new task would require retraining for. It covers the geometry tasks only, where Ultralytics spans detection, segmentation and depth together.

• https://raw.githubusercontent.com/ByteDance-Seed/Depth-Anything-3/main/README.md recorded 2026-09-27

  README: DA3 "predicts spatially consistent geometry from arbitrary visual inputs, with or without known camera poses"; capabilities "Monocular Depth Estimation", "Pose-Conditioned Depth Estimation", "Camera Pose Estimation"; "DA3 Metric Series ... fine-tuned for metric depth estimation in monocular settings".