Download YOLOE weights

How to load YOLOE weights

pip install ultralytics

from ultralytics import YOLOE

# Text-prompt open-vocabulary detection + segmentation.
# get_text_pe() encodes the class names with MobileCLIP (downloaded on first use).
model = YOLOE("yoloe-11l-seg.pt")
names = ["person", "bus"]
model.set_classes(names, model.get_text_pe(names))
results = model.predict("image.jpg")

# Or run a prompt-free checkpoint with its built-in vocabulary (no MobileCLIP):
# model = YOLOE("yoloe-11l-seg-pf.pt")
# results = model.predict("image.jpg")

Text/visual prompt vs prompt-free: which YOLOE do you want?

Every YOLOE size ships in two modes. Pick the mode first, then the size. The seg checkpoints do detection and instance segmentation from the same file.

1. 1You want to detect your own classes by typing their names

   These take class names, or a visual example, at runtime via set_classes(). This is the open-vocabulary mode, and text prompts need the MobileCLIP text encoder below.

   A text/visual-prompt checkpoint (e.g. yoloe-11l-seg.pt)
2. 2You want good zero-setup detection with no prompt at all

   The -pf variants carry a large built-in vocabulary and run straight out of the box, with no prompt and no MobileCLIP download.

   A prompt-free checkpoint (the -pf files)
3. 3You are unsure which backbone to pick

   The YOLO11 and YOLO26 backbones are the most efficient here. YOLOE-11L posts 35.2 LVIS mAP at 26.2M params; YOLOE26-L pushes that to 36.8 LVIS mAP. The v8 variants are there for the older ecosystem.

   Start with an 11-series (yoloe-11l-seg.pt) or the newer 26-series

Published accuracy exists mainly for the large sizes (YOLOE-11L: 52.6 COCO / 35.2 LVIS mAP, 26.2M params; YOLOE26-L: 36.8 LVIS mAP, 32.3M; YOLOE26-S: 29.9 LVIS mAP). The smaller sizes download and run the same way; the authors just did not publish a full per-size table, so those rows list no number rather than a guessed one.

About YOLOE

More model weights to download

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

• YOLO26 weights
• YOLO11 weights
• YOLO12 weights
• YOLOv10 weights
• YOLOv9 weights
• YOLOv8 weights
• YOLOv5 (v5u) weights
• RT-DETR weights
• YOLO-World weights
• YOLO26-seg weights
• YOLO26-pose weights
• YOLO26-obb weights
• YOLO26-cls weights
• SAM 2.1 weights
• SAM weights
• MobileSAM weights
• FastSAM weights
• Depth Anything V2 weights

See all 19 model families →

Frequently asked questions

Do I need MobileCLIP to run YOLOE?

Only for text prompts. When you pass your own class names (via set_classes / get_text_pe), YOLOE encodes them with a MobileCLIP text encoder (mobileclip_blt.ts), which Ultralytics downloads automatically the first time. The prompt-free (-pf) checkpoints use a built-in vocabulary and need no MobileCLIP. Grab the MobileCLIP file directly below for offline or air-gapped installs.

What is the difference between the YOLOE prompt and prompt-free checkpoints?

The plain seg checkpoints (e.g. yoloe-11l-seg.pt) are prompt-driven: you tell them what to find with text or a visual example at runtime. The -pf checkpoints (e.g. yoloe-11l-seg-pf.pt) are prompt-free: they ship with a large fixed vocabulary and detect it with no prompt, which is simpler to deploy but not customizable at inference time.

Is YOLOE free for commercial use?

YOLOE's weights are AGPL-3.0, the same as Ultralytics YOLO: free for open-source and research, but a closed-source commercial deployment needs an Ultralytics Enterprise license. The MobileCLIP text encoder is separately MIT-licensed.