New Release: Ultralytics v8.3.122

Discussion
,
1

:rocket: Ultralytics v8.3.122 Release: Smarter CoreML Exports, Speed Estimation, Flexible Installs & More!

:glowing_star: Summary

Ultralytics v8.3.122 brings smoother CoreML support for classification models and practical improvements to speed estimation, tracking, and installations. With easier integration for Apple environments, more accurate speed metrics, detailed tracking docs, flexible dependency control, and updated compatibility, this release delivers new power and convenience to every user!

:tada: New Features & Major Improvements

CoreML Classification Model Export

  • Native Xcode Preview Support: YOLO classification models exported to CoreML now include full metadata for seamless Xcode preview and validation, making iOS/macOS development easier than ever.

Region-Free, User-Friendly Speed Estimation

  • No More Regions Needed: Simplified object speed estimation—just set meter_per_pixel instead of drawing a region, with new parameters (meter_per_pixel, max_speed, max_hist, fps) for full flexibility and higher accuracy.

Tracking & ReID Documentation Overhaul

  • Tracker Configuration Clarified: In-depth parameter tables, step-by-step ReID instructions, and improved explanations help you get the most from your trackers. BoTSORT’s default appearance threshold is stricter for robust multi-object tracking.

Flexible & Custom Installation Methods

  • Advanced Installation Options: The quickstart guide now includes instructions for installing Ultralytics without dependencies, using custom forks, or swapping out packages for special environments.

Dependency Updates

  • Future-Proof Compatibility: Removed version ceilings from streamlit and shapely dependencies, allowing you to safely use the latest package versions.

:chart_increasing: Why You’ll Love v8.3.122

  • For Apple Developers: Preview YOLO classification models straight in Xcode for a supercharged iOS/macOS workflow.
  • For Speed-Critical Applications: Simplified, more reliable speed estimation—perfect for traffic, retail, sports analytics, and beyond.
  • For Tracking Power-Users: Detailed docs and improved defaults make tracking and ReID setup faster and more robust.
  • For System Admins & Power Users: Advanced install options accommodate even the most customized environments, such as headless servers and enterprise-grade systems.
  • For Everyone: Stay compatible with the latest Python ecosystem for smoother upgrades and fewer headaches.

:books: Full Changelog & How to Upgrade

See the complete list of changes in the 8.3.122 changelog.

Ready to get started? Update via pip:

pip install -U ultralytics

Explore the release and join the discussion at v8.3.122 Release Page.

:speaker_high_volume: We Want Your Feedback!

We thrive on your contributions and insights. If you try out the new features or run into any issues, please share your experience—your feedback drives the direction of future releases!

Thank you to the Ultralytics community and contributors for your incredible support and collaboration.

Happy YOLOing! :tada:

2

Thanks for your update.
So this v8 model is better than yolov11 or 12?

3

Hi CODE_MASTER,

Thanks for reaching out!

Ultralytics YOLO11 represents our latest state-of-the-art models, generally offering improved performance over YOLOv8. You can find a detailed technical comparison in our documentation comparing YOLOv8 vs YOLO11.

YOLOv8 remains a highly capable and versatile model, known for its balance of speed, accuracy, and ease of use within a mature ecosystem, as detailed on the YOLOv8 model page.

We do not recommend using YOLO12 due to training instability, slower speeds, and higher memory requirements. For the best performance and stability, YOLO11 is the preferred choice.