Ultralytics Release v8.3.153: Enhanced OpenVINO Benchmarks, Accurate Edge Device Results, Better Tracking, and More!
Summary
We’re excited to announce Ultralytics v8.3.153! This release brings major improvements for deploying and benchmarking YOLO11 models—especially on the latest Intel® Core™ Ultra™ 7 265K hardware—plus documented and reproducible edge device benchmarks, simplified multi-video tracking, and a fix for per-class metric reporting. Whether you’re working on laptops, powerful desktops, or resource-constrained edge devices, this update helps you get accurate, actionable insights from your vision workflows.
New Features & Updates
OpenVINO Benchmarks on Intel Core Ultra 7 265K
- Added detailed benchmark tables and charts for YOLO11 (n, s, m, l, x) models across GPU, CPU, and NPU, under both FP32 and INT8 precisions.
- Empower your deployment choices with up-to-date performance data for real-world Intel chips.
- Author: @ambitious-octopus
See pull request #20877
Raspberry Pi & 
Rockchip Docs Refresh
- Updated benchmark results for YOLO11 on Raspberry Pi and Rockchip RKNN, matching latest COCO128 dataset runs.
- Clearer, more actionable instructions to help you set up and test on these devices.
- Author: @lakshanthad
Raspberry Pi PR #20990
Rockchip RKNN PR #20991
Tracking Example Improvements
- Simpler and more robust tracking example code and documentation, specifically for YOLO11 multi-video tracking.
- Get started faster and customize for your applications.
- Author: @RizwanMunawar
See pull request #20986
Improvements & Bug Fixes
Per-Class Metrics Reporting Fix
- Corrected issues with class name alignment in detection, segmentation, and pose metric summaries.
- Now all per-class results are accurate and clearly labeled for trustworthy evaluation.
- Author: @Laughing-q
See pull request #20995
Minor Documentation Fixes
- Fixed a dataset naming typo in
rockchip-rknn.md(PR #21006 by @RizwanMunawar) - Various clarity and formatting improvements throughout the documentation for a smoother user experience.
Purpose & Impact
- Choose the Best Hardware: With fresh OpenVINO and edge benchmarks, you know exactly how YOLO11 will perform on your target device—no guesswork.
- Reproduce and Compare Results: Consistent benchmarks and clear instructions mean your results can match ours, or provide a strong starting point for your own performance tests.
- Build Powerful Video Pipelines: Updated tracking examples help you implement multi-object and multi-video tracking with less friction.
- Trust Your Metrics: Per-class results are now accurate—making evaluation and comparison of models straightforward and reliable.
- Smarter Workflows for All: Improved docs benefit both newcomers and seasoned vision practitioners.
Try It Out & Give Feedback!
We encourage you to download the new release, experiment with the updated benchmarks and tracking examples, and share your feedback or questions right here in the forum. Your experience and input help the entire community grow and improve!
Full changelog: v8.3.153 release notes
Thanks for being part of the Ultralytics and YOLO community—your feedback, code, and ideas drive these improvements. We hope you enjoy this release and look forward to your thoughts!