Ultralytics YOLO27:
Coming soon

Ultralytics YOLO27 models

Faster, more accurate, and simpler YOLO models. YOLO27l is the first Ultralytics model past 60 mAP, and every size improves, with the same one-line swap.

Ultralytics YOLO27 logo on the brand gradient

Trusted by industry leaders

DuolingoShellSiemensRenaultPhilipsNEURA RoboticsMercado LibreTata SteelFlock SafetyIntelDefense Intelligence AgencyDHL
DuolingoShellSiemensRenaultPhilipsNEURA RoboticsMercado LibreTata SteelFlock SafetyIntelDefense Intelligence AgencyDHL
DuolingoShellSiemensRenaultPhilipsNEURA RoboticsMercado LibreTata SteelFlock SafetyIntelDefense Intelligence AgencyDHL

What's new in Ultralytics YOLO27

Simpler model lineup

Four models, four clear answers. n and s for edge devices, m for production GPU workloads, l for accuracy-critical applications. No x model to evaluate as YOLO27l already delivers flagship accuracy.

Hybrid architecture

The best architecture for each model size, rather than one design stretched across the whole range. Small models stay fast on limited hardware; large models go further on accuracy.

Stronger accuracy

YOLO27l is the first Ultralytics model past 60 mAP on COCO, with higher accuracy across the range and strong accuracy-speed tradeoffs.

Seven tasks

Object detection, instance segmentation, semantic segmentation, depth estimation, image classification, pose estimation, and oriented object detection. 28 models across four sizes, all sharing the same interface.

Ultralytics YOLO models at a glance

Compare recent Ultralytics YOLO model families across speed, accuracy, and supported vision tasks.

FeaturesUltralytics YOLOv5Ultralytics YOLOv8Ultralytics YOLO11Ultralytics YOLO26Ultralytics YOLO27
Speed
Image processing time
0.54 ms0.47 ms0.56 ms0.67 ms0.62 ms
Accuracy
mAP50-95
34.3%37.3%39.5%40.9%42.3%
Object detection
Instance segmentation-
Semantic segmentation--
Depth estimation---
Image classification-
Pose estimation-
OBB detection-

Why join the waitlist?

Ultralytics YOLO27 is coming soon. Sign up once and be the first to hear.

Product updates

Architecture details, benchmarks, and release notes as we publish them.

Launch notification

Get notified when YOLO27 is available to download, train, and deploy.

Industry examples

See how YOLO27 fits manufacturing, logistics, security, and aerial inspection.
Join waitlist
Security & compliance

Enterprise-grade security, certified

Independently certified and audited, so your data stays protected in transit and at rest.

Join waitlistView security details

SOC 2 Type I

Attested against AICPA Security, Availability, and Confidentiality TSC

ISO 27001:2022

Certified information security management system

Encryption

Data encrypted in transit and at rest, across all storage and transport

GDPR & CCPA

Aligned with global data protection regulations

Why choose Ultralytics YOLO27?

Proven at scale

Ultralytics YOLO serves billions of inference jobs a day across manufacturing, logistics, security, and aerial use cases. You're adopting a model family that's already running in production.

Deploy anywhere you already work

Over 20 export formats and close to 50 integrations, built with the hardware vendors themselves. The model you train is the model you ship.

Clear commercial licensing

Flexible options for academic, open-source, and commercial use under AGPL-3.0 and Enterprise. Know where you stand before you build.

Built with the community that uses it

More than a thousand contributors test, report, and improve these models across real production deployments.

YOLO27 model family preview

Detection uses compact CNNs for N/S and query-based, NMS-free models for M/L; other tasks use CNNs. These preliminary benchmarks may change. Weights and package support are not available yet: the example is for use after release. N/S detection defaults to NMS; nms=False selects the NMS-free head used for timing, while N/S accuracy depends on the selected head.

# After YOLO27 is released
pip install -U ultralytics
from ultralytics import YOLO

# After YOLO27 weights and package support are released
model = YOLO("yolo27n.pt")
results = model("image.jpg", nms=False)
ModelSize (px)mAP 50-95CPU ONNX (ms)RTX PRO 6000 (ms)Params (M)FLOPs (B)
YOLO27n64042.316.1 ± 1.60.62 ± 0.003.07.2
YOLO27s64049.633.2 ± 0.10.79 ± 0.0011.828.2
YOLO27m64055.867.2 ± 0.31.39 ± 0.0022.865.0
YOLO27l64060.4149.6 ± 1.42.32 ± 0.0072.3165.3

Preliminary Ultralytics Platform results on COCO val at 640 px. CPU: EPYC 9655, ONNX FP32. GPU: RTX PRO 6000, TensorRT 11 FP16. Params/FLOPs use fused models. See YOLO27 docs for details.

Explore YOLO27 on ultralytics.comRead the YOLO27 technical docs

Frequently asked questions

  • YOLO27 is the upcoming Ultralytics model generation. The announced family includes four sizes and seven vision tasks. Explore the official YOLO27 overview for product details and the documentation for technical information.

  • The announced lineup has four sizes: YOLO27n and YOLO27s target edge devices, YOLO27m targets production GPU workloads, and YOLO27l targets applications that prioritize accuracy. The hybrid architecture adapts to each size, and Ultralytics reports YOLO27l exceeds 60 mAP on COCO. Compare the announced improvements, then validate models on your own data when released.

  • The YOLO27 overview announces object detection, instance segmentation, semantic segmentation, depth estimation, image classification, pose estimation, and oriented object detection. Four sizes across seven tasks make a planned family of 28 models with a shared interface.

  • Organize your images, review annotations, and establish a baseline with YOLO26 today. Use Ultralytics Platform to prepare datasets and training workflows, then evaluate YOLO27 on the same data when it launches.

  • Join the official YOLO27 waitlist for product updates and a launch notification. Use the released YOLO26 models for current projects while YOLO27 is coming soon.

Join the Ultralytics YOLO27 waitlist

Get product updates and a launch notification when YOLO27 is ready.

Join the YOLO27 waitlist