Ultralytics

Ultralytics

Ultralytics offers the YOLO family of real-time computer vision models plus a cloud platform to annotate, train, and deploy vision AI from cloud to edge.

Follow
Tags
ai-image-recognitionAI API
Last Updated Oct 21, 2026
Ultralytics screenshot

About

Ultralytics builds real-time computer vision tools centered on its YOLO model family, which the company describes as widely used for tasks such as object detection, segmentation, and image classification. Developers and enterprises can license YOLO models for their own products or run an end-to-end workflow — annotating data, training models on cloud GPUs, and deploying them globally — on the Ultralytics Platform. The models cover seven vision tasks within a single model family and are designed to run everywhere from large cloud infrastructure to low-power edge devices.

The company pairs an open-source codebase (over 1.2K contributors and 138.8K+ GitHub stars, per its own site) with commercial licensing and a hosted platform, so teams can start free under AGPL-3.0 and move to an Enterprise License for closed-source commercial use. Ultralytics reports billions of YOLO usages per day across customers in manufacturing, security, healthcare, retail, and logistics.

Key Features

Seven vision tasks, one model family: A single YOLO model line covers tasks including detection, segmentation, and classification, simplifying model selection across projects.
Ultralytics Platform: Annotate with SAM-powered and YOLO Smart annotation, train on cloud GPUs, and deploy to managed global regions without maintaining infrastructure.
26 cloud GPU types: Rented by the hour with no minimums or markup, ranging from Ampere cards at $0.25/hr to Blackwell B300 at $7.39/hr.
22 export formats: Models can be exported to ONNX, TensorRT, CoreML, and other formats for on-premise, edge, or self-managed cloud deployment.
Agents workflow builder: Chain YOLO models, logic, and actions such as Slack alerts into visual workflows.
AutoTrain: Automated training on the platform that reduces the manual work of tuning model runs.
Ultralytics Lite: An open-source desktop workspace that integrates with coding assistants like Claude Code, Codex, and Gemini.

Pricing & Fees

  • Pricing: Free plan at $0 (with $25 one-time credits, 100 GB storage); Pro at $29 per seat/month (or $30/seat/month billed yearly is higher — yearly billing saves 17%); Enterprise is custom-priced. Cloud GPUs start at $0.24/hr.
  • Free tier: The Free plan includes 3 concurrent cloud training jobs, 3 cloud deployments, 100 models, and export to all 22 formats under the AGPL-3.0 license.
  • Licensing: Free and Pro plans use AGPL-3.0; closed-source commercial use requires the Enterprise License, which includes on-premise deployment options, SSO/SAML, and custom SLAs on top of ISO 27001 and SOC 2 compliance.
  • Upcoming model: Ultralytics YOLO27 is in final R&D, planned in four sizes across seven vision tasks, with no launch date set.

Who It's For

Ultralytics suits machine learning engineers and developers building production computer vision systems, from defect detection on factory lines to safety monitoring on job sites. Industrial automation teams use it for quality control and PPE compliance; robotics builders use it for navigation and perception. The case studies listed on its site span mining conveyor inspection, rail data collection, underwater monitoring, food inspection, and sports analytics, indicating broad applicability for any team needing real-time image understanding on cloud or edge hardware.

FAQ

What is Ultralytics?Ultralytics is a computer vision company that develops the YOLO family of real-time vision models and a cloud platform for annotating data, training models, and deploying them across global regions.

Is there a free plan?Yes, the Free plan costs $0 and includes 100 GB of storage, 3 concurrent cloud training jobs, 3 cloud deployments, and model export to all 22 formats, distributed under the AGPL-3.0 license.

Can Ultralytics YOLO be used commercially?Commercial closed-source use requires an Enterprise License, which removes AGPL-3.0 obligations and adds dedicated support, custom SLAs, and on-premise deployment options.

Which platforms and formats does it support?Models can run on-premise, at the edge, or in a user's own cloud, and export to 22 formats including ONNX, TensorRT, and CoreML; the platform manages deployments across 42 global regions.

What GPUs are available for cloud training?The platform offers 26 NVIDIA GPU types by the hour with no markup or minimums, from RTX 2000 Ada at $0.24/hr to Blackwell B300 at $7.39/hr.