YOLO26: YOLO Model for Real-Time Vision AI [2026]
Solutions
Resources
Pricing<br>Docs<br>Blog
Search
Sign In<br>Book a demo<br>Get Started
Search
Sign in<br>Book a demo<br>Get Started
Search
Blog<br>What is YOLO26? An Introduction
Contributing Writer
Published<br>Jan 14, 2026<br>5 min read
SUMMARY
YOLO26 is an end-to-end object detection and multi-task model family supporting detection, instance segmentation, pose estimation, oriented object detection, and image classification across five size variants from Nano to Extra Large. Released in January 2026, it removes Non-Maximum Suppression for lower latency and drops the Distribution Focal Loss module for better compatibility with edge and low-power hardware. This post covers the architecture, COCO benchmark results, download links, and comparisons to models including RF-DETR, LW-DETR, and D-FINE.
YOLO models are a family of real-time computer vision models designed to handle a wide range of tasks, including object detection, segmentation, pose estimation, classification, and oriented object detection.<br>Leveraging popular architectures, these models offer exceptional speed and accuracy, making them well-suited for applications across edge devices, cloud APIs, and more.<br>In this blog, we’ll examine YOLO26, released January 2026, revealing its key improvements, important features, and how it compares to other leading computer vision models.<br>💡<br>Roboflow supports YOLO26 for labeling, training, and deployment, learn more.
What Is YOLO26?<br>YOLO26 is a multi-task model family designed to handle a broad range of computer vision tasks, including object detection, instance segmentation, image classification, pose estimation, and oriented object detection. The lineup features multiple size variants Nano (N), Small (S), Medium (M), Large (L), and Extra Large (X) to cater to different performance and deployment needs.<br>Compared to previous YOLO generations, YOLO26 is optimized for edge deployment, featuring faster CPU inference, a more compact model design, and a simplified architecture for improved compatibility across diverse hardware environments. Notable improvements include decreased latency by removing NMS and results staying consistent in fp16 and fp32, making it possible to run the model in an optimized, low-latency way and get the same high accuracy you saw during training.<br>RF-DETR Neural Architecture Search (NAS) is faster and more accurate than YOLO26. Read the blog post here.
Try YOLO26 on Images<br>See how YOLO26 performs on images for common objects included in the COCO dataset. Test out how the model handles your data below.
Download YOLO26<br>The table below provides links to download YOLO26 for object detection and outlines the Ultralytics reported performance benchmarks for the YOLO26 model family, comparing variants from nano to extra-large across key metrics like accuracy (mAP), latency, and computational cost.
Modelsize(pixels)mAPval50-95SpeedCPU ONNX(ms)SpeedT4 TensorRT10(ms)params(M)FLOPs(B)YOLO26n64040.938.9 ± 0.71.7 ± 0.02.45.4YOLO26s64048.687.2 ± 0.92.5 ± 0.09.520.7YOLO26m64053.1220.0 ± 1.44.7 ± 0.120.468.2YOLO26l64055.0286.2 ± 2.06.2 ± 0.224.886.4YOLO26x64057.5525.8 ± 4.011.8 ± 0.255.7193.9
This comparison highlights the trade-offs between inference speed and detection precision, enabling you to select the optimal model size for your specific hardware constraints. For other model task types, visit YOLO26 Github.<br>YOLO26 Architecture<br>YOLO26 introduces several major improvements including:<br>Broader Device Support: It removes the Distribution Focal Loss (DFL) module, simplifying inference, enabling multiple export formats (TFLite, CoreML, OpenVINO, TensorRT, and ONNX), and broadening support for edge and low-power devices.<br>Enhanced Small-Object Recognition: It utilizes the ProgLoss and STAL loss functions, improving detection accuracy, particularly for small objects, and providing significant advantages for IoT, robotics, and aerial imagery applications.<br>End-to-End Predictions: It eliminates Non-Maximum Suppression (NMS) as a post-processing step, producing predictions directly to reduce latency and make deployment in real-world systems faster, lighter, and more reliable.<br>Faster CPU Inference: Optimizations in model design and training make YOLO26 faster on CPUs compared to YOLO11. For instance, the YOLO26-N variant delivers up to 43% faster CPU inference than the YOLO11-N, making YOLO26 ideal for real-time performance on devices without a GPU.<br>Improved Training: It introduces the MuSGD optimizer, a hybrid of SGD and Muon inspired by Kimi K2 LLM breakthroughs, ensuring stable training and faster convergence by transferring optimization advances from large language models to computer vision.<br>YOLO26 Alternatives<br>Besides YOLO26, several other multi-task computer vision models are actively used and benchmarked on the object detection leaderboard.<br>RF-DETR<br>RF-DETR, developed by Roboflow and released in March 2025, is a family of real-time detection models that support...