Ultralytics YOLO27: Join the waitlist.
License YOLO
Ultralytics Inference

YOLO inference, wherever you build

Run vision AI in native applications and in the browser with one Rust-powered inference package. No Python server required.

Try the demoView on GitHub

One engine. More places to run.

Build with the same YOLO tasks across native applications and the web.

Native Rust

Run ONNX models on CPU or supported accelerators, with images, video, and live streams.

In the browser

Use WebGPU when available, with an automatic CPU fallback and no inference server.

Seven vision tasks

Detect, segment, classify, estimate depth and pose, or work with oriented boxes.
Try it live

See YOLO26 in your browser

Choose a task and model. Images and webcam stay on your device; the model downloads on first use.

Built for speed

Fast inference, close to your data

The Rust pipeline handles preprocessing and results while optimized runtimes execute the model on your device.

9.3 ms
YOLO26n detection on WebGPU
24.2 ms
YOLO26n detection on CPU

Mean model inference time after warmup for FP32 640×640 YOLO26n detection, measured on Apple M4 with Chrome 153, @ultralytics/yolo 0.0.48, and LiteRT.js 2.5.3. The demo’s first run and other devices may be slower. See benchmark details.

Choose your package

Start in Rust or bring YOLO to your web application.

Rust

Native YOLO inference with ONNX Runtime, a CLI, and accelerator support.
cargo add ultralytics-inference
View on crates.io

Web (npm)

YOLO inference in JavaScript and TypeScript, powered by Rust and WebAssembly.
npm install @ultralytics/yolo @litertjs/core
View on npm
Get started

Build with Ultralytics Inference

Explore the docs, run the examples, and bring YOLO to your next project.

Read the docsView on GitHub