Introducing Ultralytics YOLOv8

Our latest release in the YOLO family of architectures, Ultralytics YOLOv8 is a state-of-the-art model for real-time object detection, segmentation, and classification. YOLOv8 has well-documented workflows, clean code written from the ground up, easy-to-use models, and flexible solutions to fit every user's needs, with support for other YOLO versions. (YOLOv8 has since been succeeded by newer models such as Ultralytics YOLO26.)

For us, Ultralytics YOLOv8 represents the culmination of years of research, development, and testing; and our team has worked hard to ensure that YOLOv8 is not only feature-rich but intuitive and easy to use. Whether you are a seasoned professional or new to the field, we’ve created YOLOv8 so that anyone can get started with our open-source tool.

And as we continue to improve Ultralytics YOLOv8, we encourage your suggestions, PRs, bug fixes, and issues. For professional support, please fill out our contact form. As an open-source tool, the community’s feedback is vital to us. Our community has been our greatest asset and why we can keep creating and innovating.

So without further ado, let’s dive into Ultralytics YOLOv8!

What Makes YOLOv8, YOLOv8?#

We’re excited to claim Ultralytics YOLOv8 as the latest release in the YOLO family of architectures. It’s an honor to be a part of a community that has put in countless hours and effort to create models that are so widely used.

Ultralytics YOLOv8 is the next in line in the YOLO family due to building on the successes of previous YOLO versions. We’ve included many new features and improvements, which significantly increase performance and flexibility. One key aspect of YOLOv8 is its extensibility, as it is designed as a framework that is compatible with all previous versions of YOLO. This allows users to easily switch between different versions and compare their performance, making YOLOv8 a suitable choice for those who want to take advantage of the latest YOLO technology while still utilizing their existing YOLO models.

Ultralytics YOLOv5 vs. YOLOv8#

We've taken note of the demand for accessible AI tools, and folded this into our R&D process. And with simplicity and usability at the forefront, it's no secret that Ultralytics YOLOv5 has experienced widespread adoption. However, YOLOv8 takes this even further.

What’s new?#

• Ultralytics YOLOv8 is State-of-the-Art. Although YOLOv5 was fast, easy, and accurate, YOLOv8 goes further: at launch, it offered a better balance of speed and accuracy than other available real-time models.
• Ultralytics YOLOv8 is even simpler. With YOLOv5, it was necessary to clone the repo and set up your environment manually. While this is still an option for those who enjoy getting their hands dirty, to get started with YOLOv8, you can install it as a pip package.
• Ultralytics YOLOv8 becomes a platform. Migrating away from a single architecture, YOLOv8 is designed as a framework that supports multiple YOLO versions, including those developed by other teams.
• Ultralytics YOLOv8 will have a paper. A point of contention with YOLOv5 was always the lack of scientific paper. With YOLOv8, expect a paper to be published in the coming weeks.

What’s the same?#

• Clearly-documented workflows. With Ultralytics YOLOv8, we’ve maintained the documented workflows that have proven valuable in YOLOv5. However, with YOLOv8 we’ve folded improvements into them to make them even simpler.
• The creators. If you haven’t gotten it by now, no worries! Our team here at Ultralytics created both YOLOv5 and YOLOv8.
• The integrations. Ultralytics YOLOv8 will also continue supporting our partners' tools.

What else can we expect?#

The next two big releases with Ultralytics YOLOv8 that you can expect are:

• The paper. We expect to publish the scientific paper shortly after Ultralytics YOLOv8 goes live. This time, we promise.
• Performance tracking with Ultralytics HUB. We are working on a direct, tight integration between YOLOv8 and Ultralytics HUB, which will automatically allow users to visualize their model metrics and losses, preview predictions, and compare against other models, public or private. In Ultralytics HUB (now Ultralytics Platform), YOLOv8 models can be exported to many formats, including ONNX, OpenVINO, TensorRT, TFLite, CoreML, and many more.

Be sure to check out the new Ultralytics YOLOv8 repo and give us a star! You can also try the official YOLOv8 models on Ultralytics Platform.