Ultralytics launches end-to-end vision AI platform

Ultralytics has launched a software platform for computer vision projects that combines annotation, model training, export and deployment in a single workflow, as companies face growing pressure to show returns on artificial intelligence spending.

The move expands the London-based group beyond its YOLO family of object detection models into the tools used to build and run vision systems in production. Before the formal launch, developers had already uploaded more than 50 million images and created more than 150 million annotations on the platform.

The product push comes as many businesses struggle to move AI projects from pilots into live operations. Citing McKinsey research, Ultralytics said 88% of companies now use AI regularly, but only 21% have reached production scale with measurable returns. In computer vision, projects often depend on separate tools for data labelling, model training, experiment tracking and deployment, slowing development and adding operational friction.

Ultralytics is best known for the YOLO model family, which is widely used for real-time object detection. According to the company, those models account for 2.5 billion daily inferences. It also cited 125,000 GitHub stars and more than 225 million Python package downloads as signs of developer adoption.

Unified workflow

The new platform centres on a connected pipeline covering the main stages of computer vision work. Annotation tools sit alongside cloud training, model export, deployment management and monitoring.

For dataset creation, the platform includes smart annotation features built with Segment Anything Model, or SAM. Users can generate masks, bounding boxes and oriented boxes. It supports the five main vision tasks identified by Ultralytics and accepts YOLO and COCO dataset formats, as well as raw image and video uploads and community datasets.

Training is available in the cloud across 22 GPU options or on local hardware, with real-time metric streaming included. The platform also stores checkpoints and organises experiment runs through dashboards showing confusion matrices, precision-recall curves and side-by-side metrics.

Deployment is available across 43 regions through dedicated auto-scaling endpoints. Models can also be exported in 17 formats, including ONNX, TensorRT, CoreML, TFLite and OpenVINO, covering cloud, mobile, edge and embedded systems.

Model roots

Ultralytics is positioning the platform around a simple argument: the team that built the underlying models also built the surrounding software stack. That could matter in a market where many vision platforms rely on third-party models and integrations.

According to Ultralytics, YOLO26, YOLO11 and earlier versions of the model family are native to the platform architecture rather than added through external connections. That design shapes the annotation editor, training pipeline and deployment endpoints, and reflects the group's experience of how models train, export across hardware environments and behave in production.

The approach is intended to address a long-standing gap in the vision AI market. While object detection models have improved rapidly, many development teams still manage handoffs between multiple tools and services, creating delays between a successful prototype and a stable production system.

Founder and CEO Glenn Jocher framed the launch around that issue.

"Most computer vision projects today never make it past the pilot stage, not because the models aren't good enough, but because the path from experiment to production is still too complex," said Glenn Jocher, Founder and CEO, Ultralytics. "We built the Ultralytics Platform to make that path simpler. One platform, from first label to live endpoint."

Paula Derrenger, VP of Growth, described the product as a response to fragmentation in the current tooling market.

"We didn't set out to build another annotation tool or another training service," said Derrenger. "We built the platform that should have existed from the beginning. It's the only end-to-end vision AI platform native to the world's most deployed object detection models. A platform designed around how vision AI actually moves from idea to production."

Commercial plans

The platform is available through a free plan with signup credits and access to annotation, training, export and deployment. Paid Pro plans add more compute, storage and team collaboration features. Custom-priced plans are available for larger organisations with higher capacity requirements.

The launch puts Ultralytics in more direct competition with providers of machine learning operations software, annotation services and hosted deployment platforms. Its advantage may depend on whether YOLO's large user base adopts an integrated toolchain from the same supplier rather than continuing with mixed software stacks.

For Ultralytics, the platform marks a shift from being known mainly for open-source model development to selling a broader production environment for vision AI teams, with deployment infrastructure and monitoring now part of the offering.