Model weights · RT-DETR

Download RT-DETR weights

A transformer detector built for real time: an efficient hybrid encoder replaces hand-tuned anchors and NMS, matching YOLO speed at higher accuracy under a permissive license. Pick a size below and download the official .pt checkpoint in one click, with published COCO accuracy, size and license all in view.

Apache-2.0AGPL-3.0 via UltralyticsBaidu · 2023 · Ultralytics (PyTorch)

RT-DETR checkpoints, pick a size and download

A transformer detector built for real time: an efficient hybrid encoder replaces hand-tuned anchors and NMS, matching YOLO speed at higher accuracy under a permissive license.

Using it with the Ultralytics package?

The RT-DETR weights are Apache-2.0, but the Ultralytics package that most people load them with is AGPL-3.0. Running RT-DETR through Ultralytics puts your own project under AGPL-3.0 too, which means publishing your source if you distribute it or offer it over a network. An Ultralytics Enterprise license removes that. The Apache-2.0 terms only cover you if you run the weights outside the Ultralytics package.

  • RT-DETR-L
    rtdetr-l.pt
    COCO mAP
    53.0
    Params
    32M
  • RT-DETR-X
    rtdetr-x.pt
    COCO mAP
    54.8
    Params
    67M

Scores are COCO mAP at 640px, published by the authors. Weights host: github.com. Clicking Download verifies the file and starts it straight from the official CDN.

How to load RT-DETR weights

Install the Ultralytics package, then point the loader at the checkpoint. It downloads automatically on first use, or you can pass the local path to the file you downloaded above.

pip install ultralytics
from ultralytics import RTDETR

# Downloads on first use, or pass a local path to your .pt file
model = RTDETR("rtdetr-l.pt")
results = model("image.jpg")

Want to see the architecture? Export to ONNX and open it in the ONNX visualizer for a labelled diagram with shapes and parameter counts.

Open visualizer →

Choosing a RT-DETR checkpoint

Read off the table above: what each step up the size ladder costs and returns for this family specifically, where it stops being worth it, and what the license actually permits.

What the size ladder buys

RT-DETR spans RT-DETR-L at 32.0M parameters to RT-DETR-X at 67.0M, 2.1× the model, for 1.8 more COCO mAP (53.0 to 54.8). That works out to roughly 0.05 points per additional million parameters across the whole range, across the family's only step. Published figures from Baidu at 640px, not re-measured here.

Where it sits in the catalog

Against the 8 families here that publish COCO mAP, RT-DETR's best checkpoint ranks 4th at 54.8, against 57.5 for YOLO26. Read that as a tier indicator rather than a head-to-head: the families use different size ladders, input sizes and training recipes, and each number is the authors' own. The closest alternatives for object detection are YOLO26, YOLO11 and YOLO12.

Licensing, in practice

The weights are Apache-2.0. More importantly, the usual way to run RT-DETR is through the `ultralytics` package, which is AGPL-3.0, and that reaches your application even though the weights themselves are not. A permissive badge on the weights is not permission to ship closed-source over that path. Apache-2.0 on the weights is commercial-friendly and needs only attribution.

About RT-DETR

RT-DETR (Baidu) is a real-time detection transformer. An efficient hybrid encoder and IoU-aware query selection give it end-to-end, NMS-free inference with accuracy above comparable YOLO models, at the cost of a heavier backbone. It is integrated into Ultralytics with permissive Apache-2.0 licensing, making it a strong choice when license terms matter.

Author
Baidu
Released
2023
Tasks
Detect
Framework
Ultralytics (PyTorch)
Input size
640px
License
Apache-2.0
Resources
DocsGitHubPaperLicense

More model weights to download

Each page lists every checkpoint with accuracy, parameters, license and a one-click download.

See all 19 model families →

YOLO tracker FPS benchmarks →Get in touch →
Blog

From the blog

Tutorials, code, and notes on computer vision, deep learning, and applied AI.

View all posts →

Frequently asked questions

How do I download RT-DETR weights?
Click the Download button next to any variant in the table above; we verify the file and start it straight from the official CDN. You can also let the loader fetch it automatically on first use with `model = RTDETR("rtdetr-l.pt")`. All 2 RT-DETR checkpoints are hosted officially.
Is RT-DETR free for commercial use?
RT-DETR is released under Apache-2.0. Apache-2.0 is permissive and commercial-friendly: you can use it in closed-source products with attribution. Always confirm against the linked license text.
Which RT-DETR model size should I use?
Start with RT-DETR-L, the smallest and fastest, ideal for prototyping, edge and CPU. Move up the ladder only when you need more accuracy and have the compute for it. The largest variant reaches 54.8 COCO mAP.