📝 SummarXiv

Abstract

This study presents Key Architectural Enhancements and Performance Benchmarking of Ultralytics YOLO26 for real-time edge object detection, providing a comprehensive overview of the design principles of YOLO26, technological advances, and deployment readiness. YOLO26, released in September 2025 by Ultralytics, represents the newest and most cutting-edge member of the You Only Look Once (YOLO) family, engineered to push the boundaries of efficiency and accuracy on edge and low-power devices. This paper highlights architectural innovations in YOLO26, including end-to-end NMS-free inference, removal of Distribution Focal Loss (DFL) for streamlined exports, introduction of ProgLoss and Small-Target-Aware Label Assignment (STAL) for improved stability and small-object detection, and the adoption of the MuSGD optimizer inspired by large language model training. In addition, we report performance benchmarks for YOLO26 across edge devices, specifically NVIDIA Orin Jetson platforms, and compare results against YOLOv8 and YOLO11 (previous Ultralytics releases) as well as YOLOv12 and YOLOv13, which bridged the lineage between YOLO11 and YOLO26. Our comparative analysis highlights superior efficiency of YOLO26, accuracy, and deployment versatility, establishing it as a pivotal milestone in the YOLO evolution.