Ultralytics Launches YOLO26, Setting a New Global Standard for Edge-First Vision AI

Ultralytics YOLO26 delivers world-leading end-to-end, NMS-free performance for the fastest and simplest production deployment in computer vision

LONDON--(BUSINESS WIRE)--Ultralytics, the global leader in open-source vision AI, today announced the launch of Ultralytics YOLO26, the most advanced and deployable YOLO (You Only Look Once) model to date. Engineered from the ground up for edge and low-power devices, YOLO26 introduces a fully end-to-end, NMS-free architecture that fundamentally redefines how object detection is trained, deployed, and scaled in production, delivering industry-leading performance with dramatically reduced complexity.

YOLO26 is built for demanding environments where efficiency, reliability, and hardware flexibility matter, including robotics, manufacturing automation, smart cities, logistics, healthcare, retail, and embedded AI.

Key architectural and training innovations in YOLO26 include:

• Exporting and deploying models is difficult: DFL removal improves exportability and compatibility with edge hardware.
• Post-processing slows deployment: End-to-end, NMS-free inference reduces latency and simplifies production.
• Training small or complex objects is unstable: Progressive Loss Balancing + STAL improves stability and small-object detection.
• Training can be unstable: MuSGD ensures efficient and stable training.
• CPU performance is often too slow: Up to 43% faster CPU inference allows real-time and accurate detection on existing hardware at lower cost.

Traditional object detection pipelines, built around fragile post-processing and GPU-heavy assumptions, are a hard blocker for real-world edge deployment. YOLO26 eliminates these constraints, enabling organizations to deploy state-of-the-art vision AI where it actually runs: on CPUs, edge accelerators, and embedded devices.

“YOLO26 is the fastest and most deployable object detection system available in the world today,” said Glenn Jocher, Founder and CEO of Ultralytics. “By removing entire stages of the inference pipeline, YOLO26 delivers state-of-the-art performance without the complexity that has held edge deployment back for years.”

Ultralytics YOLO26: Built for Real-World Edge Deployment

YOLO26 represents a shift toward native end-to-end object detection, producing predictions directly without post-processing steps such as NMS. This design significantly reduces latency, removes fragile post-processing steps, and shortens time-to-production by simplifying integration across cloud, edge, and embedded environments.

“YOLO26 represents a breakthrough in end-to-end object detection,” said Jing Qiu, Author of Ultralytics YOLO26 and Senior Machine Learning Engineer at Ultralytics. “By eliminating NMS and simplifying the model architecture, we achieve faster inference, stronger stability, and unmatched hardware portability, without sacrificing accuracy.”

A Unified Model Family Across Vision Tasks

YOLO26 is released as a multi-task model family, supporting object detection, instance segmentation, classification, pose estimation, and oriented object detection within a single, unified framework. Each model variant is designed to support training, validation, inference, and export, enabling teams to move from research to production faster and with fewer compromises.

“Ultralytics YOLO26 proves that world-class AI and real-world deployability no longer have to be a trade-off,” said Paula Derrenger, VP of Growth at Ultralytics. “This release removes the friction that has historically slowed down production vision AI, allowing teams to deploy faster, scale confidently, and standardize on a single best-in-class model family from cloud to edge.”

Ultralytics YOLOE-26: Open-Vocabulary Segmentation Models Built on YOLO26

Ultralytics is also introducing YOLOE-26, a new family of open-vocabulary segmentation models built on the latest YOLO26 architecture and training innovations.

YOLOE-26 is not a feature or a new task, but a specialized model family that reuses YOLO26’s end-to-end design to enable text prompts, visual prompts, and prompt-free segmentation. Available across all standard YOLO sizes, YOLOE-26 delivers stronger accuracy and more reliable real-world performance than previous open-vocabulary segmentation models, while integrating seamlessly into existing segmentation workflows.

Continuing the Open-Source Ultralytics YOLO Legacy

YOLO26 builds on Ultralytics’ long-standing commitment to open-source innovation. YOLO models are trusted by millions of developers worldwide and are used 2.5 billion times per day across industries and geographies. By continuously evolving YOLO with a focus on real-world constraints, Ultralytics enables organizations to deploy vision AI at scale, without sacrificing performance or transparency.

Through partnerships with innovators such as Axelera AI, Intel, DEEPX, and Sony AITRIOS, Ultralytics enables YOLO models to run efficiently on advanced AI hardware platforms. Together with a broad ecosystem of platform and hardware partners, Ultralytics continues its mission to deliver accessible, high-impact vision AI from cloud to edge.

YOLO26 will be available through the Ultralytics platform, with full support across training, inference, and export workflows. Enterprise licensing options are available for organizations deploying YOLO26 in commercial and closed environments, with support for production deployment, long-term maintenance, and scalable edge rollouts.

For more information and access to Ultralytics YOLO26, visit: platform.ultralytics.com/ultralytics/yolo26

About Ultralytics

Founded by Glenn Jocher, Ultralytics is the leading force in vision AI, best known for its Ultralytics YOLO (You Only Look Once) models. With 123,000 GitHub stars, 205+ million Python package downloads, and close to 2.5 billion daily usages, Ultralytics YOLOv5, YOLOv8, YOLO11, and now YOLO26 have become widely recognized object detection models globally.

Ultralytics empowers developers and enterprises with easy-to-use, high-performance vision AI technology. Its mission is to simplify and democratize AI, making it accessible and impactful across industries, including manufacturing, agriculture, healthcare, and logistics.